Battery Energy Storage Systems for the New Electricity Market Landscape: Modeling, State Diagnostics, Management, and Viability—A Review

Abstract

Renewable energy penetration and distributed generation are key for the transition towards more sustainable societies, but they impose a substantial challenge in terms of matching generation with demand due to the intermittent and unpredictable nature of some of these renewable energy sources. Thus, the role of energy storage in today’s and future electricity markets is undisputed. Batteries stand out among the different alternatives for energy storage. The R&D effort into different battery chemistries contributes to reducing the investment associated with battery systems. However, optimizing their operation according to the users’ and the electricity markets’ needs is the turning point to finally make these systems attractive. This review delves into the topic of battery management systems from a battery-technology-independent perspective, and it also explores more fundamental but related aspects, such as battery modeling or state estimation. The techno-economic part of battery energy storage systems is also covered in this document to understand their real potential and viability.

1. Introduction

The European Union (EU) has set the goal of becoming climate-neutral by 2050. In the European Green Deal [1], the European Commission (EC) proposes a roadmap to promote a resource-efficient economy with net zero greenhouse gas emissions by 2050. To achieve this objective, the energy system will undergo a major transformation towards an electrified, decentralized, intelligent, and flexible system where more than 80% of electricity will be obtained from renewable energy sources [2]. In turn, in the REPowerEU Plan [3], the EC has proposed increasing the target for renewable energy sources to 45% by 2030, which will contribute to quickly replacing fossil fuels and accelerating the transition to clean energy in Europe. The EU’s electricity system will increase the proportion of renewable energies in its generation mix, from 37% in 2021 to 69% in 2030 [4]. In this context of high penetration of renewable energies, technologies that provide flexibility to the electrical system will be essential in order to match demand with generation. These technologies include energy storage, demand response, supply flexibility, and interconnections. Energy storage stands out due to its capacity to provide services at different scales and timeframes as a function of the technology [5].

Nowadays, pumped hydroelectric power plants constitute the predominant storage technology worldwide, with 160 GW installed in 2021 and a capacity of 8500 GWh in 2020 (over 90% of the total electricity storage). On the other hand, batteries are experiencing significant development, with a 60% increase (over 6 GW) in capacity compared to 2021, with the US (2.9 GW), China (1.9 GW), and Europe (1.0 GW) leading the market. Although their current power, 16 GW, is much lower than pumped hydroelectric energy, a rapid increase in battery capacity is expected in the coming years. The International Energy Agency estimates in its Net Zero Scenario that 680 GW of grid-scale battery storage will be required by 2030 [6]. In Europe, studies conducted by the EC indicate that to achieve decarbonization goals, 480 GW of flexibility will be required by 2030, 67 GW of which will be stationary batteries [7]. The Asia Pacific region currently leads the way in battery development and manufacturing, serving as the largest market and exhibiting the fastest growth projected in the coming years, largely driven by the automotive sector [8]. Li-ion battery technology dominates the market, and it experienced a 65% increase in demand by 2022 compared to 2021 [9]; China holds the largest share of global lithium and cobalt refining and is a leader in graphite production [10,11]. Even if other regions, such as North America and Europe, are heavily investing in cell production capacities and attempting to diversify supply, they will still rely on China [12] to achieve their decarbonization goals, which are strongly linked to renewables and storage.

A battery management system (BMS) is a key element in monitoring and controlling the operation of a battery energy storage system (BESS). Its functionalities include ensuring operational safety, enhancing power delivery reliability, and increasing BESS performance and battery life, among other tasks. Various components, such as modeling and state estimations, including the state of charge (SoC), state of health (SoH), and aging prediction, are fundamental parts of the software in a BMS. They play a crucial role in managing the batteries based on the monitored parameters. Numerous reviews have recently addressed these topics of modeling, state estimation, and BMSs, with the focus being predominantly on lithium-ion batteries, as demonstrated in

Table 1. Table containing previous reviews on the topics of battery modelling, state estimation, and BMS.

Topic	Article	Observations
Battery modeling SoC, SoH, and RUL estimation	Ren and Du (2023) [13]	ML for SoC and SoH. LIBs for EVs.
	Zhang et al. (2023) [14]	Data-driven SoH of LIBs. Accurate and sufficient datasets are crucial for appropriate predictions.
	Shao et al. (2023) [15]	RUL prediction methods, LIBs.
	Ghalkhani and Habibi (2023) [16]	AI and ML for more accurate prediction of SoC, SoH, etc. Focus on EVs. Also on battery design, thermal management.
	Ul Hassan et al. (2022) [17]	SoC estimation techniques for batteries in power grids, including a detailed description of each and highlighting their pros and cons in power grid applications. Reliability in estimation crucial to maximize renewable energy sources utilization and to support the grid.
	Luo et al. (2022) [18]	Deep learning for SoH, SoC, and RUL prediction for LIBs.
	Hossain Lipu et al. (2022) [19]	Deep learning for SoH, SoC, and RUL prediction for EVs.
	Ansari et al. (2022) [20]	RUL prediction methods for LIBs. Identification of issues and challenges in RUL prediction and research gaps.
	Wang et al. (2022) [21]	Battery state prediction. Accurate models are key for ensuring reliability and optimized operation. Data-driven methods based on AI have large potential.
	Rouholamini et al. (2022) [22]	In one of its sections, this article covers battery modeling.
	Puleston et al. (2022) [23]	Comprehensive study of dynamic models for VRFB available in the literature, together with an analysis of the existing model-based estimation strategies.
	Cui et al. (2022) [24]	SoC definition, SoC estimation of LIBs using neural network methods, methods classification, error metrics evaluation for different approaches, and makes recommendations for next-generation BMSs.
	Jiang and Song (2022) [25]	SoH estimation methods for lead–acid batteries.
	Chan et al. (2022) [26]	Electrochemical impedance spectroscopy SoH estimation for LIBs.
	Peng et al. (2022) [27]	Capacity estimation methods for BMSs in EVs and RESs and provides practical and feasible advice for capacity estimation with onboard BMSs.
	Wang et al. (2021) [28]	Deep-learning for RUL prediction for LIBs.
	Espedal et al. (2021) [29]	SoC estimation for LIBs in EVs.
	Basia et al. (2021) [30]	SoH estimation techniques and presents a novel feature-based classification.
	How et al. (2019) [31]	SoC estimation methods highlighting the model-based and data-driven approaches.
	Meng et al. (2018) [32]	Battery modeling methods with potential to be used in a model-based SoC estimation structure.
	Madani et al. (2018) [33]	Equivalent circuit models and parameters identification methods in lithium-ion batteries for energy storage applications.
BMS	Liu et al. (2023) [34]	Conventional battery management, challenges, and data-driven approach.
	See et al. (2022) [35]	BMS integration; recommendations for safety design and performance optimization.
	Krishna et al. (2022) [36]	Traditional BMSs have limitations, and digitalization to improve BESS performance.
	Tran et al. (2022) [37]	Cloud-based BMS for LIBs. Traditional BMS limited by low computational capacity and data storage; solved by cloud capabilities.
	Gabbar et al. (2021) [38]	BMS and standardization for transportation and stationary. Recommendations for BMS development based on several aspects.

AI = artificial intelligence, BESS = battery energy storage system, BMS = battery management system, EV = electric vehicle, LIB = lithium-ion battery, ML = machine learning, RES = renewable energy system, RUL = remaining-useful-life, SoC = state of charge, SoH = state of health.

.

Given the expanding market of BESSs, there is an increasing need to develop battery technologies. These should either be producible using locally available resources or ensure rapid adaptation to available technologies under competitive conditions. This urgency is felt in various regions, including the European Union, among others [39]. Consequently, there is a justification for developing a battery management system (BMS) that is capable of efficiently accommodating any of the available battery technologies, as well as the battery modeling and battery state estimation associated with these [40,41]. As mentioned above, and to the best of our knowledge, most of the focus in the literature reviews so far has been devoted to lithium-ion batteries. However, this review aims to explore the topic of battery management, modelling, and state estimations from a broader perspective for the various battery technologies available, and to identify both commonalities and specific characteristics, with special attention to stationary applications.

The structure of the paper is represented in Figure 1. After the contextualization and justification of the need for this review article, dealt with in Section 1, Section 2 follows, which is devoted to battery modeling and state estimation (SoC, SoH). Section 3 addresses the BMS and important aspects, such as charging and discharging, cell balancing, and functionalities. The techno-economic perspective of BESSs is also reviewed in Section 4, focusing on the services and applications that batteries can fulfill. The metrics used to evaluate their profitability in various works within the literature are also examined.

2. Battery Modeling and Predictions

2.1. Battery Modeling

Generally, a battery cell is an electrochemical complex system with multiple non-linearities that escalates as cells are connected, conforming to a battery pack. These non-linearity effects are due to multiple factors:

-

Polarization effects depend on the state of charge; this phenomenon is related to battery internal resistance [42,43].

-

Hysteresis effects, resulting in many possible SoC values for one voltage value. These effects are noticeable in lead–acid, nickel–metal hydride, nickel–cadmium [44], and Li-ion batteries [45]. It is important in Li-ion batteries to take this effect into account due to its flat open-circuit voltage curve, as not considering it may lead to an important state of charge estimation error.

-

Diffusion and dynamic effects, which are due to ionic flow.

-

Parasitical reactions, resulting in charge loss, especially during charging [46].

-

Non-linear dependency of the open-circuit voltage curve, polarization effects with temperature, and total available capacity with temperatures; this is particularly true for cold temperatures [43].

In addition, battery internal states, which are fundamental to safe and efficient operation, are not directly measurable. The main purpose of the models in the context of battery management systems is to obtain more accurate estimates of these internal states. It is also important to consider that aging mechanisms are highly dependent on non-measurable states, such as SoC.

The following classification of battery models is generally accepted in the literature [47,48,49]:

-

Physics-based models: Derived from fundamental electrochemical and conservation laws (charge, mass, energy), these models provide high accuracy but are computationally complex. They are generally used offline for battery design and characterization [45,47,50,51]. These models provide insights to internal battery mechanisms, and can also be used to generate algorithms for SoH estimation [52,53]. Their complexity can be reduced, leading to reduced-order models that can be implemented in a BMS [48].

-

Equivalent circuit models (ECMs): Empirical models that substitute the battery for an equivalent circuit whose electrical components, generally resistances, capacitors, and voltage sources, are fitted to laboratory data to try to obtain the best representation of the particular battery modelled. This review will focus on ECMs, as they are commonly used in BMSs due to their low computational cost and ease of implementation.

-

Mathematical models: These models define useful relationships between battery parameters but do not try to represent a circuit or link the model parameters to battery internal mechanisms necessarily [47]. One example of these models is the Shepherd model [48].

▪

Data-driven models: Highly popular due to data availability, accuracy, and ease of implementation. Several data-driven algorithms are being considered for battery modeling, SoC, and SoH; these are typically Decision Trees, support vector machines, Artificial Neural Networks, and Markov chain models. Some examples can be found in references [48,54].

While different technologies have different characteristics, modeling approaches remain similar because the fundamental electrochemical mechanisms are the same. Generally, equivalent circuit models are the preferred option to adopt in a BMS.

2.1.1. Equivalent Circuit Models

One model that can be seen as a reference and that is the basis for most equivalent circuit models used nowadays is the Randles circuit (Figure 2a) [55]. The resistor R_s models the electrolyte resistance, the double layer capacitance C_dl represents the effect of charges accumulating in the electrolyte at the electrode’s surface, R_ct models the voltage drop due to charge transfer, and Z_w is a Warburg impedance, a type of electrical element that models diffusion in a dielectric. In order to represent Warburg impedance with common electrical components, an infinite number of capacitor-resistor pairs that would yield the same transfer function would be needed. As this is impractical, an approach with one to three pairs is generally used and provides sufficient accuracy. The number of RC pairs gives name to the order of approximation of the Randles circuit.

It is a common practice to omit the effects of double layer capacitance, as it has a low impact except on very high frequencies. Electrolyte resistance and charge transfer resistance are usually joined together. With all of these simplifications, the resulting circuit is shown in Figure 2b. If a voltage source is added to this circuit, a Thevenin model is obtained. This voltage source represents the open-circuit voltage of the cell (OCV), and its value depends on the state of charge value and the temperature. If the model is further simplified and RC pairs are omitted, then a R_int model is obtained. If hysteresis effects are added to the Thevenin model, an enhanced self-correcting cell model is obtained [45]. It is also common to add a parallel resistor to the open-circuit voltage source to represent battery self-discharge.

These models can be enhanced arbitrarily, modeling the dependence of each parameter with temperature and SoC. This dependence can be expressed in terms of a lookup table (LUT) or an equation fit. More parameters and dependencies require more experimental data and complex parameter fitting strategies. Some authors explore the trade-off between the increased complexity of a model and its associated errors [56,57,58]; this is important, as models should be as simple as possible while maintaining good accuracy.

Another type of equivalent circuit model, the impedance-based model, emerges from the results of electrochemical impedance spectroscopy (EIS) [47]. EIS obtains information about battery impedance behavior for different frequencies. This impedance relationship with frequency is usually represented in a Nyquist plot, as in Figure 3. From this analysis, impedance-based models fit a complicated equivalent network to mimic battery impedance spectra.

Regarding various applications for BESSs, it is important to note that using representative current profiles of the final use is fundamental for obtaining circuit parameter values that ensure good model accuracy. What constitutes good accuracy depends on the specific application.

2.1.2. Equivalent Circuit Models for Different Technologies

Table 2. Equivalent circuit models for different battery technologies.

Reference	Technology	Model Type	Observations
Salameh et al. (1992) [59]	Lead–acid	Thevenin model modified	Substitutes voltage source for a capacitor. One RC pair used. Different resistors used in charge and discharge. Self-discharge considered. Temperature compensation for self-discharge resistor. Battery pack model. No error metrics provided.
Ceraolo (2000) [46]	Lead–acid	Thevenin model with self-discharge	Two RC pairs used. Thermal effects modelled. Self-discharge considered. Battery model. No error metrics provided.
Chen et al. (2006) [47]	Lead–acid, NiCd, NiMH, Li-ion, polymer Li-ion, and others	Thevenin model modified	Two RC pairs used. Voltage source on circuit modeling capacity for OCV. Battery capacity depends on cycle number and temperature. Self-discharge considered. Value of resistor dependent on cycles, SoC, and temperature. Battery model. Max voltage error 30 mV; max runtime error 0.395%.
Ng et al. (2008) [60]	Lead–acid	Thevenin model	One RC pair used. OCV values dependent on SoC and current rate. Battery model. No error metrics provided. Focus on SoC estimation.
Peng, (2011) [42]	Lead–acid	Thevenin model	One RC pair used. Thermal effects modelled. LUT parameters’ dependence with SoC and temperature. Max. error 1.49% for dynamic current profile. Battery pack model.
Roselyn et al. (2021) [44]	Lead–acid	Shepherd model	Separate branches for charge and discharge. Hysteresis considered in SoC estimation, not in terminal voltage. Battery pack model. No error metrics provided. Focus on SoC estimation.
Lavety et al. (2021) [61]	Lead–acid	Thevenin model	One RC pair used. Model of battery cell. Only discharge is modelled. Battery pack model. Max. error 0.4% for step increasing/decreasing load.
Yi-Feng et al. (2021) [62]	Sodium-ion	Thevenin model	Multiple RC pairs tested. From 0 to 3. Focuses only on discharge. Fitting of parameter dependency with SoC with polynomials. Bayesian information criteria used to set polynomial order. Battery model. RMSE error 5.5 mV for three-order model.
Rabab et al. (2022) [63]	Sodium-ion	Thevenin-model-based	Interesting insights about operating areas of sodium-ion batteries. Parameters estimation from physics equations. One RC pair used. Dependence of parameters with temperature. Battery model. Relative error at high SoC 4.2% and 20.72% for low SoC.
Norian (2011) [64]	Nickel–cadmium	Thevenin model	One RC pair used. Parameters dependence of SoC. Battery model. No error metrics provided. Focus on parameter identification.
García-Plaza et al. (2015) [65]	Nickel–cadmium	Thevenin-model-based	Two RC pairs used. Three circuits used to model battery cell: hysteresis, Ah counting, and voltage current. Self-discharge considered. Dependency of parameters with SoC. Battery pack model extrapolation from battery model. Average cell model error 0.34% and stack model error 0.44%.
Micea et al. (2011) [66]	Ni-MH	Thevenin model	One RC pair. OCV–SoC relationship updated online. Battery model. No error metrics provided. Focus on remaining cycles estimation.
Xuyun et Zechang (2008) [67]	Ni-MH	Impedance-based	One RC pair considered and a Warburg impedance. Hysteresis considered in open-circuit voltage. Battery pack model. No error metrics provided, only graphics.
Yin-Zhu (2016) [68]	Ni-MH	Impedance-based	Modified Randles circuit with an additional ZC pair. Constant phase impedance element used to model non-ideal mechanisms. Offers interpretation of electrochemical phenomena with circuit elements. Insights about behavior change with aging.
Buller et al. (2003) [69]	Li-ion, supercapacitors	Impedance-based	Zarc and Warburg elements modelled as multiple RC pairs. Charge transfer resistance modelled with Butler–Volmer equation. OCV–SoC dependency stored in LUT. No error metrics provided, only graphics.
Erdinc et al. (2009) [70]	Li-ion (not specified)	Thevenin model modified	Parameters dependent on SoC, equations. Capacity update considering cycle and calendar losses. Effects of temperature considered. Series resistance with dependency on cycles elapsed. Battery model. No error metrics provided, only graphics.
Xiong et al. (2011) [71]	LiFePo4	Thevenin model	OCV–SoC relationship from Nernst equation. One RC pair used. Battery model. Max. error less than 1% for Dynamic Stress Test current profile.
Karsten Propp et al. (2016) [72]	Li-S	Thevenin model modified	Behavioral modification of Thevenin model. Temperature effects considered. Polynomial functions for parameter dependency with temperature. Parameters dependency with SoC. Self-discharge not considered. RMSE 0.32 mV for New European Driving Cycle current profile.
Jiang et al. (2017) [73]	Li-S	Thevenin model	One RC pair used. Insights about Li-S internal mechanisms and model deviations. Battery model. Error 3% in most SoC regions, excluding lower SoC values.
Cleary et al. (2022) [56]	Solid State Li-S	Thevenin model	Multiple RC pairs studied. Series resistance dependency with SoC. Linear interpolation to model OCV and series resistance dependency with SoC. Battery model. RMS error range depending on model complexity. 9.498–18.730 mV using current profile similar to the Hybrid Pulse Power Characterization Test.
Baccouche et al. (2022) [74]	Li-NMC	Thevenin Model	One and two RC pairs used. Parameters dependency with SoC and Crate. Different parameters used for charge and discharge. Temperature effects not considered. (Constant at 25 °C). Battery model. Error less than 1% for SoC range of 10–100% using Dynamic Discharge Pulse current profile.
Sadhukhan et al. (2022) [75]	LiFePo4	Thevenin model	Two RC pairs used. Parameters dependent on SoC. OCV–SoC fitted with polynomial equation. Battery model. No error metrics provided. Focus on SoC estimation.
Karimi et al. (2023) [76]	Li-C	Thevenin model	Two RC pairs used. Temperature effects considered (−30 °C, 60 °C). LUTs for model parameters, for charge and for discharge. Battery model. Max. error 3% using dynamic current profile.
Zhang et al. (2015) [77]	VRFB	Thevenin model modified	First-order Randles circuit. Vanadium diffusion and shunt current modelled. OCV modelled with simplified Nernst equation. Temperature effects not considered. Internal series resistance dependent on SoC. LUT for parameter values, flow rates, and current densities. Hydraulic model and pump considered. Mean error 0.09 V using dynamic current step profile.
Mohamed et al. (2013) [78]	VRFB	Thevenin model	Two RC pairs used. Diffusion and shunt currents not considered. Temperature, current density, flow rate, and concentrations not considered. Extended Kalman Filter for online parameter estimation. Mean error 4.9 mV during charge–discharge cycle.
Xiong et al. (2019) [79]	VRFB	Thevenin model modified	Coupled electro-thermal model. Diffusion and shunt currents modelled through resistors. Two RC pairs used. Temperature effects considered. OCV modelled with Nernst equation. Max. error 15 mV per cell using wind generation current profile.
Khaki et al. (2021) [80]	VRFB	Thevenin model with electro-chemical model	Complex dynamic model used to determine OCV. Thevenin model with one RC pair. Adaptative Kalman Filters for parameter estimation. Mean error 0.88% for ECM-Hybrid Extended Kalman Filter and 0.79% for ECM-Particle Filete using Constant Current—Constant Voltage charge profile.
Woodfield et al. (2022) [81]	VRFB	Thevenin model modified	Same model used as Zhang et al., 2015 [77] but includes comprehensive shunt current resistors network. Includes thermal and hydraulic model. No error metrics provided; purely a simulation article.

provides a review of several articles comparing different equivalent circuit models used for various battery technologies and making relevant observations. In summary, the following can be noted:

-

Equivalent circuit models deliver sufficient accuracy, with errors in the order of mV, a wide range of applications, encompassing both static and dynamic current profiles, as well as different technologies.

-

It is essential to consider the effect of temperature, SoC, and current rate effects on battery parameters.

-

Hysteresis is not frequently studied and can lead to important errors.

-

Only a few papers address aging in battery modelling and parameter updating.

-

Typically, most papers use one or two RC pairs; a higher number of RC pairs is rarely observed, due to increased complexity they introduce.

2.2. State of Charge (SoC) Estimation

The state of charge (SoC) is a parameter that indicates the remaining charge in the battery and, therefore, the remaining energy available. It is related to the chemical species used in the redox or intercalation reactions. Estimating the SoC is crucial to prevent operation outside the safe area due to overcharge/overdischarge (which may cause permanent damage), to avoid unpredicted system interruptions, and to predict the time remaining until cell depletion. Theoretically, it can be defined as the ratio of the charge available at one moment by the maximum charge the cell can withhold in a stable manner. It can be expressed in terms of positive or negative electrode. Note that 100% SoC or 0% does not necessarily mean that the cell has the maximum or minimum theoretical concentration, as practical limitations have to be taken into consideration (such as electrode layer collapse, excessive volume changes, and conductor dissolution due to low voltage).

Two concepts should be introduced related to SoC: total capacity and available total capacity. The former is related to the charge the battery can store and is independent of the current rates applied or the temperature, while the latter refers to the charge the battery can deliver/accept at a specific moment of time given the temperature and the current rates applied. Voltage drops, which are highly related to the current rate and temperature, can limit battery operation, thereby limiting the charge delivered/accepted.

There is not a universal consensus regarding the SoC definition. The SoC is generally defined as the level of charge relative to its capacity; the problem here is the definition used for capacity. Some authors refer to the available capacity [82,83], while others refer to the total capacity [45], and others refer to the rated battery capacity [84,85,86]. Also, some authors consider the updating capacity as the battery ages [25,45].

With regard to redox flow batteries, the SoC definition has to take into account the two electrolytes used, i.e., the catholyte and anolyte. Due to imbalances and system aging, the SoC in each electrolyte can differ, although some authors assume the system is balanced and, therefore, SoC calculations for each electrolyte yield the same result [87]; however, later studies used the minimum SoC of both electrolytes [88], and others use an average value [89]. Also, in redox flow batteries, some authors have proved that it is possible to track SoC via absorbance measurement [90,91].

2.2.1. SoC Estimation Methods

In the literature, SoC estimation methods are generally classified as follows:

-

Coulomb counting methods: These methods are based on current sensing and integration. The drawbacks of these methods are generally that they are prone to error accumulation due to measurement bias and their high sensibility to the initial SoC estimate. Despite this, they might be a reasonable solution for low-precision applications where recalibrations can be made frequently.

-

OCV methods: These methods are based on the relationship of the open-circuit voltage with the SoC. The main challenge is that they require the battery to be disconnected and to rest for a sufficient time to achieve a representative OCV, which may not be possible in certain applications. In the case of Li-ion batteries, they present a very flat OCV–SoC curve; thus, a small error in the OCV measurement can lead to a considerable error in the SoC. Another aspect that has to be considered is hysteresis, as, if its effects are significant, then it is impossible to make a one-to-one correspondence between the SoC and the OCV.

-

State-observer methods: As the SoC is not a quantity directly measurable, these algorithms combine inputs and outputs estimated by models and measurements to try to obtain an estimate of the SoC as accurately as possible. Generally, these methods are based on different versions of Kalman filters (i.e., Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), Sigma-point Kalman Filter (SPKF), etc.). Also, other algorithms are used, such as the Sliding Mode Observer (SMO), High Gain Observer (HGO), and Particle Filters (PF), though the latter are generally computationally expensive. These algorithms are highly popular and allow for online applications with high accuracy.

-

Data-driven methods: These methods are based on the use of machine learning techniques, such as Support Vector Regression (SVR), Decision Trees (DT), various types of Artificial Neural Networks (ANN) (Extreme Machine Learners (ELM), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Feed-Forward Network (FFN), Cascaded Feed-forward Network (CFFN)), Fuzzy Logic (FL), etc.). Their high accuracy is at the expense of large data requirements. Depending on the quality of data used and the parameters considered, they may not generalize well to different batteries.

2.2.2. SoC Algorithms for Different Technologies

In this section, several articles treating SoC estimation for different battery technologies will be reviewed in

Table 3. SoC estimation algorithms for different battery technologies.

Reference	Technology	SoC Algorithm	Observations
Santhanapoongodi et al. (2016) [92]	Lead–acid	Terminal voltage-SoC LUT	12 V, 45 Ah battery pack tested. Different current rates considered.
Santos et al. (2017) [93]	Lead–acid	EKF	Effects of temperature considered. ECM is a second-order Thevenin model. Parameter estimation using MATLAB curve fitting toolbox. Linear interpolation SoC–OCV. 100 Ah battery pack tested.
Roselyn et al. (2021) [44]	Lead–acid	Coulomb counting with hysteresis correction	12 V, 62 Ah battery pack tested. Shepherd model used for battery. Genetic algorithm for parameter identification. Coulomb counting error reduced from 14.14% to 1.27%.
Sun et al. (2022) [94]	Lead–acid	Online sequence ELM	Adaboost for network training. SoC estimated from electrolyte density. Online incremental learning. RMSE 0.018–0.03 g/cm3.
Tiwari et al. (2018) [95]	Sodium-ion	CFFN (Three layers)	Fabricated battery Na0.7Ni0.3Mn0.59Co0.1Cu0.01O2 tested. Backpropagation for network training. Two current rates considered. Three temperatures considered. RMSE 0.94%.
Darbar et al. (2022) [96]	Sodium-ion	FFN (Two layers)	Na0.67Fe0.5Mn0.5O2 (NFM) as a cathode and Na metal as a reference electrode battery tested. Network trained for different current rates and cycles. RMSE 1.22–2.23%.
Verbrugge et al. (2004) [97]	Ni-MH	Coulomb counting filtered	12.5 Ah battery tested. Thevenin model with one RC pair. OCV modelled by Nernst equation. Parameter estimation using least square regression. Considers self-discharge. Hysteresis considered. SoC calculation from Coulomb counting SoC. SoC obtained from model OCV.
Windarko et al. (2010) [98]	Ni-MH	OCV	Sebang GMH 100 NiMH Battery rated at 1.2 V and 100 Ah. Rint model with different resistors for charge and discharge. Values dependent on SoC. Takacs hysteresis model to improve OCV–SoC relationship. Maximum error of 10%.
Xing et al. (2014) [99]	Li-ion	UKF	LiFePO4 18650 cylindrical type battery tested. Rint model; resistor value dependent on temperature. OCV–SoC–temperature LUT. Parameter fitting using least square regression. RMSE 1.1–24.63%.
Cheng et al. (2014) [100]	Li-ion	Finite difference EKF	Lithium iron phosphate LP2770102AC battery. Second-order Randles circuit. Parameter identification using non-linear least squares. RMSE of 1.74%.
Chung and Yang (2018) [101]	Li-ion	EKF	First-order Randles circuit; OCV modelled as capacitor. Parameters depend on SoC, temperature, and current rates. Average error 0.15%.
Fang et al. (2021) [102]	Li-ion	Fractional order UKF	Panasonic NCR18650B, 3.4 Ah, 3.7 V battery tested. Model with two RC pairs. State space equation with fractional order. Adaptive genetic algorithm for parameter identification. RMSE 0.7–1.0%.
Huang et al. (2021) [103]	Li-ion	EKF	Lithium iron phosphate 20 Ah battery pack tested. Online parameter identification with recursive least squares with forgetting factor. Levenberg–Marquardt modification of covariance matrix. RMS error 1.216%.
Berrueta et al. (2021) [104]	Li-ion	EKF simplified	Second life 66 Ah battery pack tested. First-order Randles circuit. Model with constant parameters. Comparison with Ampere Counting, Adaptive PF, and MATLAB’s EKF. RMSE 4.2–6.0%.
Zhang et al. (2022) [85]	Li-ion	RNN + UKF	Samsung INR 18650-20R and INR 18650-25R under various ambient temperatures; (CALCE) battery dataset. UKF refines output of RNN. Effects of temperature considered. RMSE error 1.5–3.4%.
Chen et al. (2023) [83]	Li-ion	EKF	2.6 Ah, 4.2 V battery tested. Parameter estimation by genetic algorithm and Levenberg–Marquardt algorithm. Second-order Randles model. MAE 0.3%.
Zheng et al. (2018) [105]	VRFB	Dual KF (EKF + KF)	5 kW/30 kWh energy storage system tested. ECM: Shunt current source modeling pump consumption. Shunt resistance modeling parasitic reactions. Series resistor. Real voltage source with parallel capacitor. KF receives output of Coulomb counting and EKF and estimates SoC. Error 1.5%.
Clemente et al. (2021) [87]	VRFB	SMO	Single-cell VRFB tested. Simplified physics model; electrolyte tanks balanced and equal flows. Particle Swarm for parameter identification.
Khaki et al. (2021) [89]	VRFB	OCV	Nine-cell VRFB prototype tested. Two models proposed: Dynamic electrochemical model. ECM (first-order Randles). SoC computed from OCV using adapted Nernst equation. RMSE error—11.19%.

.

2.3. State of Health (SoH) Estimation

The state of health (SoH) is an indicator that evaluates the aging of a battery. As batteries age, their ability to store charge decreases, their impedance tends to grow and vary in behavior, and their charge retention over time deteriorates. An accurate estimation of the SoH is crucial to prevent potential damage to the battery [106] and to predict the operational range, performance, and life expectancy of the battery [107].

A general definition of the SoH responds to the following equation [25,107,108,109,110,111,112]:

$$SoH=\frac{Q_{actual}}{Q_{initial}}$$

where $Q_{actual}$ refers to the maximum capacity of a battery for a given moment in time and $Q_{initial}$ refers to the maximum capacity at the beginning of its life, usually given by manufacturer.

Other authors use different SoH definitions considering other battery parameters, such as the resistance value parameters [113,114,115], the end of discharge voltage [116], or in terms of resistance and capacity [117].

In the case of redox flow batteries, the SoH was defined as the quotient of the minimum of moles available in the two electrolytes by the ideal amount [88,118], as the fraction of the total available concentration of electroactive species with respect to a reference value of concentration [119], or according to the net oxidation/reduction imbalance of the system [118].

2.3.1. State of Health (SoH) Estimation Methods

The taxonomy used for SoH estimation methods will be fundamentally based on the comprehensive review of [25]:

-

Direct measurement methods: This family of methods estimates SoH performing operations based on experimental measurements, such as cell voltages and currents. The great advantages of these methods are the easiness to implement them and their reduced computational complexity. On the other hand, their main drawback is the limitations that arise from obtaining these measurements.

▪

Coulomb counting methods: These methods rely on current sensing and the definition of SoH according to the equation above. Despite yielding an accurate result of SoH, they require the performance of a full charge and a full discharge to obtain the present capacity. This procedure may be difficult to implement in real applications, and its use its mainly reserved for SoH calibration in laboratories.

▪

Resistance methods: These approaches exploit the relationship between battery internal resistance and SoH. Some authors use this resistance instead of the battery capacity to estimate the SoH, as previously mentioned. These methods need to select the proper sampling frequency in order to avoid inaccurate results [109], and they consider the relationship of this resistance with the SoC. They are fast, easily implementable, and suitable for online applications; however, they have limited accuracy.

▪

EIS methods: Instead of considering a unique value for the resistance, these methods try to obtain the spectra behavior of battery impedance for a variety of frequencies (Nyquist plot). This information is then correlated with SoH data to estimate it. As mentioned before, it is important to consider the relationship with the SoC to improve the results. Despite their higher accuracy, the measurements needed require longer testing time than resistance methods, especially for low frequencies and specialized hardware.

▪

Coup de Fouet (lead–acid only): Lead–acid batteries present an under-voltage/overvoltage when they start discharging/charging, having been fully charged/discharged at a constant current rate. This phenomenon is called the Coup de Fouet. These methods pursue estimating the SoH from values of voltage peak and plateaus, which are SoC-dependent. However, the effectiveness of these methods has been questioned in the past [120].

▪

Incremental Capacity Analysis (ICA) and Differential Voltage Analysis (DVA) (Li-ion): ICA and DVA are non-invasive methods used to evaluate the SoH of batteries by tracking the electrochemical properties of the cell. ICA is based on the differentiation of the battery capacity over the battery voltage or the battery voltage derivatives, for a full or a partial cycle, regarding the experimental conditions. Several ICA research studies have been performed on various Li-ion chemistries, and several mathematical approaches have been employed to obtain the differential curves. On the other hand, DVA is the inverse technique of ICA, as it consists of studying the relationships of the voltage derivative with respect to capacity.

-

Model-based methods:

▪

Parameter-model-based methods: This approach investigates the relationship between battery model parameters and the SoH. Different types of models could be used, i.e., physical, mathematic, impedance-based, or ECM. These methods tend to produce accurate results and can be implemented in online applications. The data origin for parameter estimation may be EIS, constant current discharge, two-pulse load test, or variants.

▪

State-observer methods: These methods follow the same idea stated in the SoC section. As a matter of fact, Kalman Filters used to estimate the SoC usually estimate the battery capacity alongside the SoH.

-

Data-driven methods: These algorithms rely on large amounts of data to train models that produce accurate results. Examples of algorithms of this type are presented in the SoC section.

2.3.2. State of Health (SoH) Estimation Methods for Different Technologies

In this section, several articles dealing with SoH estimation for different battery technologies will be reviewed in

Table 4. SOH estimation algorithms for different battery technologies.

Reference	Technology	Estimation Algorithm	Observations
Shahriari and Farrokhi (2013) [107]	Lead–acid	EKF + Radial Basis Function NN	6 batteries used with different aging. 3 for training, 3 for testing. NN used for battery pack modeling and EKF for SoC estimation. Activation function of NN of Gaussian type. Online SoH estimation from charge–OCV slope, estimating the slope using recursive least squares. Constant temperature. Maximum error 3.0%.
Chaoui et al. (2015) [115]	Lead–acid	Parameter-model-based	$\mathrm{Equivalent} \mathrm{circuit} \mathrm{used} R_b+\left(C\right|\left|\left(R+OCV\right)\right)$. Genetic algorithm for parameter estimation. $\mathrm{SoH} \mathrm{calculated} \mathrm{from} R_{bat}=R+R_b$$, \mathrm{considering} \mathrm{values} \mathrm{of} R_{ba{t}_{new}}$$\mathrm{and} R_{ba{t}_{eol}}$. Linear correction of temperature with manufacturer coefficient.
Sadabadi et al. (2021) [109]	Lead–acid Starting–lighting–ignition	Parameter-model-based	Three aged battery packs tested. Second-order Randles ECM. R series value obtained from instant voltage drop. RC values from Nelder–Mead optimization. Explores evolution of ECM parameters with aging. Used a map between RC resistors, SoC and SoH. SoH estimation error 0.8–12.2%.
Haifeng et al. (2009) [114]	Li-ion	DEKF	SoH definition based on ohmic resistance value. Dual Extended Kalman Filter for ohmic resistance estimation.
Andre et al. (2013) [121]	Li-ion	DEKF + SVR	High-energy lithium-ion pouch cells with a graphite anode and a NMC cathode. Cells with nominal voltage 3.6 V and nominal capacity 10 Ah. Tested over four cells. Second-order Randles circuit. DEFK for SoC estimation and SVR for capacity estimation. Max error of SoH below 20.0%.
Fan et al. (2019) [122]	Li-ion	Parameter-model-based	Eight LiNMC battery cells with 0.94 Ah nominal capacity, 3.7 V nominal voltage tested. First-order Randles circuit. OCV–SoC–C LUT. Uses affine projection for parameter estimation. Particle swarm for capacity matching with data using OCV–SoC–C LUT. Estimates current capacity and then SoH. Relative error below 1% for all conditions and different aging states.
Lee and Lee, (2021) [112]	Li-ion	Multiple neural networks	Qualitative SoH estimation. Three values (normal, caution, fault). One multilayer for SoH diagnosis, and three NN model banks for SoC estimation. Each model bank comprises normal, caution, and fault model. (Capacity 100–90%, 90–80%, <80%). SoH reused for SoC estimation. Average error of 0.58%.
Gong et al. (2022) [123]	Li-ion	GPR, SVR, and LR	Data from MIT, CALCE, NASA, and Oxford datasets. Three energy metrics used for SoH estimation: Constant Current charge phase energy. Constant Voltage charge phase energy. Equal discharge voltage interval phase energy. Comparison of the following algorithms: Elastic net linear regression. Support Vector Regression. Gaussian Process Regression. Prediction errors 0.5%, R2 97.0%.
Ezemobi et al. (2022) [111]	Li-ion	FFNN	LiNiMnCoO2 Nominal capacity 3500 mAh, nominal voltage 3.635 V, cut-off voltage 2.5 V. Network comprised two hidden layers with 10 neuron each. Activation function for hidden layers: Hyperbolic tangent sigmoid. Activation function for output layer: Softmax. SoH binned in 5%. From 100% to 80%. SoH prediction accuracy of 96.2%.
Yang et al. (2015) [124]	Ni-MH	Non-linear regression	Commercial AA-type Ni-MH cells (Pisenr, Sichuan, China, Crated 1.8 Ah and rated voltage 1.2 V). Three charge–discharge rates. Nonlinear relationship found between the discharging capacity (Cdischarge Ah) and the voltage changes in 1 s occurring at the start of the charging process (∆Vcharge mV). Non-linear least squares method for curve fitting. The validity of the curve model verified using (Cdischarge, ∆Vcharge) data groups from the charge–discharge cycle tests at different rates. The Best (RMSE) can reach up to 1.2%.
Galeotti et al. (2015) [125]	Ni-MH	Impedance-based	Uses Theory of Evidence to determine number of cycles. Impedance at 2.5 Hz from EIS selected as representative frequency. Parameters for cycle estimation normalized voltage and imaginary impedance at 2.5 Hz for different SoC levels. Requires reference curves.
Telmoudi et al. (2020) [126]	Ni-MH	Fuzzy regression	Laboratory-made cell. Fuzzy c-regression model and Euclidean particle swarm optimization for NiMH battery. SoH determined according to the discharge rate of the battery model. Constant temperature considered.
Song et al. (2022) [127]	Single-Flow Zinc–Nickel	DUKF	Second-order Randles circuit. Parameter identification from constant current pulse tests SoC and ohmic resistance. RC parameters obtained by improved Harris hawk optimization (IHHO). Parameters are SoC-dependent. SoC, RC, and capacity parameters estimated with Dual Unscented Kalman Filter. IHHO used to obtain initial parameters of DUKF.
Clemente et al. (2023) [88]	VRFB	Parameter-model-based	Four different submodels: Electrochemical (Skyllas Kazacos), Thermal, Hydraulic, Voltage (OCV by Nernst equation considering hydrogen protons), Activation overpotential (Butler–Volmer), Ohmic overpotential (constant but different depending on whether the system is charging or discharging), Concentration overpotentials. Parameter estimation with offline estimator based on Particle Swarm optimization. SoH obtainable from model.
Puleston et al. (2023) [118]	VRFB	High-gain observer (HGO) + Dynamic inverter + selector	Skyllas Kazacos for concentrations and Nernst for OCV. Model reduction due to strong non-linearities. Dynamic of VRFB separated in two timescales. Correction terms to consider side reactions. Individual anolyte/catholyte SoC estimation and the reversible causes of the battery capacity fading are distinguished from the irreversible ones. Observability analysis performed. High gain observer + Dynamic inverter + Selector SoC and SoH estimation. Relative error below 2%. Mean 0.51%.

.

3. Battery Management System (BMS)

3.1. BMS and Functionalities

A battery management system (BMS) is responsible for monitoring and controlling a battery pack, which is the term used to describe an assembly of battery cells. The main function of a BMS is to ensure safety during charge and discharge cycles and reliability in power delivery. Secondary functions might include methods to enhance the battery’s lifespan, like cell balancing or optimization strategies to increase the system stability.

An energy storage system (ESS) typically consists of the energy storage medium (usually batteries), the BMS, and the Energy Management System (EMS), sometimes called the Power Management System (PMS), which collects the data from the BMS and external agents or sensors to the ESS (demand from power loads, status of the electric grid, data from the electric markets, etc.). Depending on the configuration, the BMS and the EMS might share or divide calculation functions. Besides the main components of the system, other subsystems may exist within the ESS, especially in large or medium applications. Relevant examples are the Power Conditioning System (PCS) (Section 3.4), intended to improve the quality of the power that is delivered to the grid or the load equipment, or the Battery Thermal Management System (BTMS), which controls the battery temperature through various active or passive cooling techniques [128]. The feedback function in the EES allows communication between these elements, especially the EMS and the BTMS. With active cooling BTMS, this communication is normally two-way to ensure a safe operation of the batteries. Specific functional areas and tasks of BMSs are outlined in

Table 5. BMS functional areas related to objectives and tasks.

Objective	Area	Description	Tasks
Safety	Main directive (Monitoring and protection of batteries)	Critical functions, which take priority	Voltage monitoring
			Temperature monitoring
Reliability			Monitoring other parameters (chemical reactions, external temperature, humidity, etc.)
			Current control
			Stop battery operation
Increase battery life	Secondary directives (Performance of the system)	Management of the storage medium	Cell balancing (on charge)
Optimize benefits			Charging cycle control
			Load balance of cells (on discharge)
Increase system stability			Isolation of cells
			Other
Faulty and status signals	Feedback (Communication)	Interrelation with other elements inside the ESS	Data connectivity
Stored energy and available output
Two-way communication with ESS and its subsystems (BTMS, etc.)

.

3.1.1. Safety and Reliability

This function is achieved in two ways: by monitoring key parameters of the battery pack or individual cells (voltage, temperature, current, etc.) and by adjusting the battery operating conditions based on these data. The BMS detects risks, like overheating, overcharging, overcurrent, overvoltage, and short-circuit, and takes action to mitigate these risks by modifying the charging current and the voltage; furthermore, when active cooling is employed via a BTMS, it issues commands to adjust the cooling process.

The Safe Area of Operation

The BMS ensures that batteries remain within the Safe Operation Area (SOA), often represented by a voltage vs. temperature graph provided by the manufacturer. When a battery voltage falls beneath the SOA, it will get overdischarged, which may reduce the life expectancy of a battery; meanwhile, when rising above the SOA, it will be on overload, which may cause an explosion on some chemistries, like the Li-ion family. Similarly, when a battery temperature goes over the SOA, it may lead to a thermal runway, which is a fire hazard, while a temperature beneath the SOA facilitates cell degradation. In conclusion, a battery working mildly outside the SOA will reduce its life expectancy, while going further off the safe range will cause damage to the batteries or risks to equipment and people potentially affected by fire or explosion.

The SOA varies by battery type and chemistry, and the relationship between voltage and temperature may be nonlinear. Figure 4 represents an SOA of a lithium-ion battery.

Table 6. Acceptable range of use of several battery technologies.

Technology	Temperature Range (Recommended)	Voltage Range (Cell)	Recommended Operation (Cell)	Notes
Lead–acid—VRLA [129]	−20 to 45 °C	2.20 to 2.35 V 1	5 to 35 °C 2.25 to 2.30 V	Using the battery for prolonged periods over 40 °C causes a thermal runway.
Lead–acid—VLA [130]	−40 to 55 °C	2.15 to 2.45 V 1		Using the battery for prolonged periods over 55 °C causes a thermal runway.
Nickel–cadmium (Ni-Cd) [131]	0 °C to 45 °C (Charge) −20 °C to 65 °C (Discharge)	0.90 to 1.50 V	1.3 V–20 °C	Charge acceptance at 45 °C is 70%. Charge acceptance at 60 °C is 45%.
Nickel–metal hydride (NiMH) [132,133]		0.90 to 1.50 V	1.2 V–20 °C
Lithium-ion (Li-ion) [134]	0 °C to 45 °C (Charge) –20 °C to 60 °C (Discharge)	~2 to 4.2 V	0–20 °C Voltage depends on chemistry	No charge permitted below freezing point. Good charge/discharge performance at higher temperature but shorter life.
Lithium–sulfur (LiS) [135,136]	−60 °C to +70 °C	1.90 to 2.45 V	2.45 V at 20 °C (Charge) 1.9 V at 20 °C (Discharge)	Storage temperature: −60 °C to 85 °C (30 °C recommended).
Lithium–metal–polymer (LMP) [137]	−30 °C to +65 °C	2.5 to 4.5 V	?
Sodium-ion (Na-ion) [138,139]	−20 °C to +60 °C	1.5 to 4.3 V	?
Sodium–sulfur (NaS) [140]	260–350 °C	1.78 to 2.208 V	?
Sodium–nickel–chloride (Na/NiCl2) [141]	270 °C–350 °C	2.67 to 3.1 V	2.58 V (nominal voltage)
Redox flow batteries [142]	?	1.15–1.55 V	?	30 °C recommended operation temperature
Zinc–bromine flow batteries (ZnBr-FBs) [143]	20–50 °C	1–1.82 V	1.5 V (nominal at ambient temperature)

Parameters for which conclusive evidence could not be found are marked as “?”. ¹—Voltage range depends on operation temperature.

summarizes the SOA of several battery types, based on the manufacturers’ data and reviews.

The C-Rate

The BMS must monitor the current during the operation of the battery pack, as each cell has a maximum current level determined by the surrounding temperature, the temperature passing through the cells, and safety considerations. This charge and discharge current is often expressed by the C-Rate, reflecting the speed of charge and discharge relative to the capacity of a battery (in Amperes divided by Amperes per hour). As shown in

Table 7. C-rate vs. duration of charge/discharge.

C-Rate	Time
5C	12 min
2C	30 min
1C	1 h
0.5C or C/2	2 h
0.2C or C/5	5 h
0.1C or C/10	10 h
0.05C or C/20	20 h

, the C-Rate affects battery duration. Varying the charging speeds can prolong battery life, and different charging methods (Section 3.3.8) have implications for life expectancy and charging rate. Though extensive data exist on battery life expectancy concerning technology and charging profiles, many BMS manufacturers use proprietary charging tables. These may adjust the C-Rate based on certain factors, such as the battery’s SoC or age, which are often aligned with the manufacturer’s specific recommendations.

Protection

The BMS monitors parameters like the current, voltage, and temperature to detect and predict potential risks during the battery’s operation. This detection process involves considering both the current state of the battery (nowcasting) and predicting its future state (forecasting). These measurements guide secondary functions, such as cell balancing, current control, and cell isolation. The BMS ensures safety through three levels of actions:

• First level: The BMS maintains cells within the SOA through cell balancing and temperature and voltage monitoring, and it may prompt active cooling action if available (BTMS takes action).
• Second level: Power absorption during charging or drawing during discharging is limited to keep the battery within safe parameters.
• Third level: The BMS halts operation if the temperature or voltage exceeds safe limits, thereby avoiding potential damage or hazards.

3.1.2. Management and Diagnosing

The BMS aims to extend battery life through a combination of diagnosis and management functions.

The diagnosis function involves forecasting the conditions of the cells that form the battery packs, with parameters like SoH and SoC playing a crucial role.

The management function calculates the current energy level stored in the battery, the power available to be drawn from the battery by the load at a given time, the estimated time for a full charge or discharge, and several other parameters required by the EMS to oversee the entire system. This function is closely intertwined with the diagnosis function and is overruled by security functions. Its main role is to regulate the charging and discharging processes of the batteries to extend their life expectancy by using various methods, such as cell balancing, current limiting, and disconnecting at specific charging conditions, like the cut-off voltage.

The operation of the management function depends on parameters like the battery technology, the SoH, and the application. Manufacturers may adjust the charging or discharging profiles based on certain factors, such as the required charge or discharge time, ambient conditions, the number of charge and discharge cycles the battery has experienced, response times of the application, and more. This could result in several charging and discharging profiles being used throughout the battery’s life or being actively adjusted based on measured conditions.

Lastly, the management function might also adopt approaches beyond optimizing battery life, such as economic considerations, availability, or grid support, thereby affecting decisions related to the battery’s operation.

3.1.3. Communication

The BMS is designed to communicate with the EMS and, optionally, with the BTMS. In certain topologies, the BMS may directly communicate with other BMSs and electronic devices, like smart chargers. If not, the connection is through the EMS.

The BMS uses a communication protocol consisting mainly of standardized solutions, such as CAN, Modbus, and serial communications, like RS-232 or RS-485. While wired solutions are typically used for reliability, wireless options, like Bluetooth, Zigbee, Wi-Fi, and cloud-based IOT communication, may also be available but are generally secondary [144,145].

In the event of a battery malfunction, the BMS will send a warning to the EMS and, optionally, other devices. Commonly, these warnings are given under two thresholds:

-

Warning: The operation conditions are outside or near the normal working parameters without exceeding the safety conditions. A warning may not prompt drastic actions from the BMS, and it may wait for feedback from the EMS.

-

Alert: The operation conditions are approaching the safety threshold limit, and the BMS will take appropriate action to ensure safety by terminating the battery operation or drastically reducing its output.

3.2. BMS Architecture and Topologies

When defining the ESS, a factor that takes importance is the way the BMS will interact with the energy storage. Though in this article we are focusing on a complete BMS incorporating the global functions, the battery packs may incorporate one or several BMS circuits to allow for protection of the cells and security. Some examples are described below:

-

Charger: Controls individual charging of a battery cell or a stack of cells in series, focusing on the charging voltage and the cell(s) temperature (e.g., [146]).

-

Protector: Protects a single cell or a stack of cells in series from overcharge or overdischarge (e.g., [147]).

-

Regulator: Balances the cells during charging and protects them against overdischarge (e.g., [148]).

-

Balancer: Incorporates monitoring and balancing functions and normally applies to several cells from the pack (e.g., [149]).

-

BMS: The ESS can incorporate several BMSs depending on distributed BMS topologies, where the BMSs will communicate with each other. Battery packs can be modular and may include just the protection system or the complete BMS with slave mode capabilities (e.g., [150]).

The connection to the BMS may follow several topologies and fall into two main categories:

-

Centralized Architecture: This involves a single BMS with a control unit that manages all the cells in one or several battery packs through multiple communication channels. It is typically cheaper but limits the distance between battery packs and the BMS, which is possibly unsuitable for large-scale applications. An issue in a pack may affect the whole system, rendering the whole ESS inoperative.

-

Distributed Architecture: Several BMSs or charge controller systems may be used while one main BMS will control the global operation. This architecture has two typical schemes:

○

Master–Slave: Each battery pack has a BMS with limited functions (slave BMS or S-BMS), while a master or M-BMS will control the whole operation and determine the charging strategies. Costlier and more robust than centralized options, this design fits better with larger systems and offers a shorter response time. The main vulnerability is that a failure in the main ESS will affect the whole pack, though the S-BMS may be able to retain partial functionality.

○

Modular: Uses several BMSs through the ESS, with one chosen as a master controlling the whole operation. This system is the sturdiest, because a failure on the main master can be fixed by promoting one of the slaves as the master, though it is the costliest.

It is also important to highlight that these topologies may also exist inside a singular battery pack at a smaller level (intra-battery communications [151]) and that combinations of these topologies into hybrid configurations are also possible [152,153].

3.3. BMS and Battery Charging

3.3.1. Battery Packs and Cells

When talking about the BMS, one important point to take into account is the battery pack topology, which is commonly named as a battery and composed of cells. A cell is the basic electrochemical unit used to generate or store electrical energy from stored chemical energy. A battery is an association of cells related to its final use; for example, in small electronics, a battery may consist of one cell, while in bigger applications, it consists of an association of two or more cells connected to each other either in series, in parallel, or in a combination of both. When a battery consists of two or more cells, it can also be called a battery pack. The battery pack arrangement follows several criteria that will be explained in further detail in following sections; the main criteria are the needed voltage, current and capacity required for the specific application.

3.3.2. The Life Cycle of a Battery

Manufacturers normally rate batteries for a number of cycles of charge and discharge until the end of its life (EOL). This metric is calculated by a drop of capacity, and a common value is 20% reduction (80% of the rated capacity), though this will differ depending on the end application [20,154,155,156]. A battery that is determined to be at its EOL for certain applications can be repurposed for less demanding applications, becoming “second life batteries”.

The number of cycles is calculated under ideal conditions, determined at a nominal voltage, ambient temperature, and maximum drawn current. However, research has identified various parameters that can influence battery life expectancy. Even though most of these parameters are mentioned in their respective sections, they have been grouped under this section for clarity:

-

Operating outside of the SOA and protections: This can have a detrimental effect on battery durability. Monitoring different parameters, such as temperatures and voltages, allows for protections to be triggered to safeguard the battery. The safety and reliability functions of the BMS address these aspects and prevent situations that might degrade the battery.

-

Charging and discharging profile of a battery (Section 3.3.8): This can impact battery life expectancy, especially if a fast charge is maintained when the battery is almost fully charged. Charging and discharging are controlled by the management functionality of the BMS. It is vital to consider that lifespan maximization may not always be the BMS optimization goal, with economics or other variables factored into the decision-making process.

-

Correct cell balancing (Section 3.3.6): Proper balancing is crucial for achieving a sufficiently long life expectancy for the battery pack. The management functionality of the BMS should oversee cell balancing, equalizing the charge between cells, and preventing premature aging.

3.3.3. The Depth of Discharge

The Depth of Discharge (DoD) is a measure of the energy drawn from the battery, calculated as the discharged current from a fully charged battery divided by its nominal capacity, and expressed as a percentage of the battery’s overall capacity. The DoD is relevant in two key aspects:

1. DoD over time: A large current drawn from the battery in a small period of time can accelerate its aging. The DoD in relation to the time of discharge is a factor to take into account by the BMS to calculate the aging of a battery. When used this way, the parameter is normally measured as Ah.

2. Maximum DoD: Depending on the battery technology, it may be advisable to keep the SoC over a certain percentage to increase the battery life. In a battery, the DoD is the complement of the SoC, and their sum equals the total rated capacity of the battery. As the battery ages and its maximum SoC declines, the maximum achievable DoD must be adjusted. In this context, it reflects the energy available to be drawn from the battery.

3.3.4. Cut-off Voltage, Self-Discharge, and Maximum Voltage

The cut-off voltage, the point at which a battery discharge is terminated, is a manufacturer-prescribed value dependent on the battery technology. The BMS controls this critical parameter. Operating below the recommended voltage threshold can damage the battery, decrease its lifespan, and boost its rate of discharge. The cut-off voltage set up by the BMS will be higher than the undervoltage limit indicated by the cell manufacturer by a security margin to ensure that the battery does not reach a deep discharge stage.

The self-discharge of a battery is the recoverable capacity loss when no current is drawn by a load, and it varies with the technology and environment. Lead–acid batteries have a rapid self-discharge rate and require voltage maintenance to preserve capacity, while others, like lithium-ion, have a much slower rate. Self-discharge is often measured as a percentage of the rated capacity lost monthly at a certain temperature. The self-discharged energy, contrary to capacity loss through aging or other factors, is recoverable during the next charge cycle, provided that the battery does not reach a deep discharge stage; this is another factor to be supervised by the BMS.

Lastly, the maximum voltage of a battery, determined by its technology and defined by manufacturers, is a crucial factor to be considered by the BMS. A cell increases its resistance to the passing of current when the voltage is near a maximum point, at which point the absorbed current goes down to low C-rates. Exceeding the maximum voltage increases the temperature of the cell, promoting its degradation and aging, with adverse effects, like thermal rundown in lead–acid technologies and lithium plating on Li-ion batteries. Aged Li-ion batteries may experience plating below the maximum rated voltage, which brings the complex task to the BMS of adjusting the charging profile to compensate for the aging of a battery [157].

3.3.5. Battery Pack Topologies

The battery pack topology refers to the arrangement of the battery cells, which is crucial for delivering the required energy to a given load or application. This topology not only determines the energy in terms of voltage and current but also greatly influences how a BMS handles cell balancing. In stationary applications, a PCS can optimize the conversion of voltage from DC to AC, among other functions. The efficiency of the PCS depends on selecting an appropriate voltage for the battery pack topology. This optimization, in turn, determines the required current that the battery pack needs to deliver to the load [158].

Thus, the following aspects must be determined:

• Maximum power the battery pack needs to deliver to the load.
• Ideal voltage to supply to the PCS.
• Maximum current the load will draw from the battery.
• Total capacity needed for the battery pack.
• Optimal topology to fulfill the requirements, extend the battery life, reduce costs, and minimize complexity. This can be achieved through various combinations of two basic topologies, series or parallel, or their hybridizations, Series Connected Configuration (SCC, Figure 5a) or Parallel Connected Configuration (PCC, Figure 5b). Information about them is summarized in

  Table 8. Comparison of different topologies.

  Topology	Cell Balancing	Characteristics	Failures and Isolation
  			Short Circuit	Open Circuit	Isolation *
  Series	Needed per cell	Same current on cells and load Load voltage, sum of cells voltages Higher current and cell temperature	Yes	Yes	Cell—reduced capacity and voltage
  Parallel	Natural equalization	Same voltage on cells and load Load current, sum of current in each cell	Yes	No	Cell—reduced capacity and maximum discharge current
  SCC	Needed per cell	Stacks of cells connected in series Stacks are connected in parallel	Yes (cell) Reduced effect (stack)	Yes (cell), disables stack No (stack)	Stack—reduced capacity and maximum discharge current
  PCC	Needed per stack (minor)	Stacks of cells connected in parallel Stacks are connected in series	Yes (reduced effect)	No (slight reduction in maximum discharge current)	Stack—reduced capacity and voltage

  * Isolation consequences for cell-stack (with/without switches).

  .

3.3.6. Cell Balancing

During the production of batteries, small variations in cell resistance, maximum voltage, and capacity can exist, even if the battery pack assembly manufacturers select cells from the same model with similar capacities. These differences grow over time due to factors like the charging speed, current levels, temperature variations, and environmental conditions, leading to disparities between the cells in the SoC, capacities, aging rates, and self-discharge rates.

Cell balancing is a critical process managed by the BMS aiming to equalize the charge between cells. An appropriate balancing technique will allow the pack to absorb more energy and deliver more instantaneous power, and it becomes more crucial as the pack ages and the differences between the cells increase.

As explained in the previous section, the series connection of cells is more prone to this issue because the parallel connection will naturally equalize the cells. However, inconsistencies between the cells of a parallel connection will still lead to a faster performance decay rate, though, granted, at a lesser level compared to a series connection, making cell balancing still necessary for parallel and hybrid topologies, especially with aged cells [159].

Cell balancing techniques can be classified into two groups: dissipative equalization techniques and non-dissipative equalization techniques.

Dissipative Equalization

Dissipative equalization techniques, known as the Cell Bypass Method (CBM) or Cell To Heat (CTH) method, dissipate extra energy from a cell using parallel elements. These techniques should be used when charging because they do not allow reverse switching, which will result in a higher imbalance during discharge [160].

-

Passive cell balancing: This method uses resistors in order to dissipate energy from the cell without active control. The main passive technique is the fixed shunt resistor method, where a fixed resistor of a calculated value is connected in parallel to allow the flow of current when a cell reaches full voltage. Other passive balancing techniques involve transistors, diodes, and variable resistors without an active control for more accurate balancing [161].

-

Active cell balancing: This approach incorporates controlled switches to actively manage energy dissipation during charging. It adds complexity due to the need for control electronics, like microcontrollers, and the need to use inputs, like the voltage, temperature, or current passing through the cells. Cost and complexity can vary, and the control scheme may be adjusted throughout the battery’s life cycle. The simplest and most common method is the switchable shunt resistor method, which entails employing control switches with the resistors in series to manage current flow on demand through each cell. Another method described in the literature is the shunt transistor method [162,163].

Non-Dissipative Equalization

These methods redistribute the energy passing through the cells to equalize the charge. These systems may include, in addition to switches, capacitors, inductors, or transformers.

-

Cell to Cell (CTC or C2C): Transfers excess energy from one cell to another using switches, capacitors, inductors, transformers (also called converters), or a combination of these. This type of balancing is divided into Adjacent Cell Balancing (ACB) and Direct Cell Balancing (DCB). In the ACB subtype, the energy flows to a cell adjacent (neighbor) to the one that is in a full state through the use of capacitors (single capacitor, switch capacitors, double-tiered switch capacitors), inductors (single inductor, multiple inductor), [160] or transformers (Ćuk converter, PWM controlled converter, quasi-resonant/resonant converter, multiple transformers) [164]. In the DCB subtype, a series of switches are introduced to control the transfer of energy between the cells and an equalizer, which may be a capacitor (flying capacitor method) [165], inductor (flying inductor method), [166] or transformer (multiphase interleaved converter) [167]. The use of switches allows for the transfer between any pair of cells on the stack or pack. In the more complex topologies, this method reaches an accurate SoC equalization between the cells, having also a good scalability and fast balancing process. The issue with this method is that an accurate control of the balancing requires the use of a big number of switches, which adds complexity to the system and cost to the BMS.

-

Cell-to-Pack Methods (CTP or C2P): Transfer energy from a higher SoC cell to the entire pack continuously and dynamically via the use of capacitors, inductors, transformers, and a monitoring circuit. CTP methods are specific to the series topology of packs, and they have their best efficiency when only one cell of the series is unbalanced with an overcharge while the rest of cells are balanced. Conversely, their worst efficiency is reached when the unbalanced cell has a lower voltage than the rest of the series. Some CTP methods are the shunt inductor method [168], the boost shunting method, the multiple transformers method [168], the switched transformer method, the multisecondary windings transformer method [169], and the time-shared flyback converter method [170]. CTP methods suffer from high switching losses and slow balancing. Though they can make use of simpler balancing structures, an accurate SoC and voltage measurement from the batteries will need to include a certain level of complexity.

-

Pack-to-Cell Methods (PTC or P2C): The energy is transferred from the whole pack to the cell with the lowest SoC in the pack, until the voltage is equalized. This type of circuit is the most adaptable for bidirectional balancing and it has slightly more efficiency than the CTP methods due to less energy losses during switching, although it has similar balancing speed. Some PTC methods are the voltage multiplier method, the full-bridge converter [171], the multiple transformers method [172], the switched transformer method [172], and the multisecondary windings transformer method [172,173].

-

Cell-to-Pack-to-Cell Methods (CTPTC or C2P2C): These are higher in complexity as they operate in several directions, from the higher charged cell to the pack (CTP), from the pack to the lower charged cell (PTC), or from Pack to Pack (PTP) when the energy is transferred between stacks in the same pack. The level of complexity adds up to the BMS cost, but it is linked to higher conversion speed and energy efficiency. Some CTPTC methods include the bidirectional multiple transformers, the bidirectional switched transformer, and the bidirectional multisecondary windings transformer [174].

3.3.7. Reconfigurable BMS

Traditionally, battery packs are organized statically, and BMS devices that make use of cells in these arrangements have been defined as a “Static BMS” by a small number of authors [175,176,177,178]. In contrast, several types of reconfigurable architectures for the battery packs have been proposed, aiming at a major level of adaptability, rapid response to different type of applications, greater reliability and safety of use, and better cell equalization and cell life cycle. Because the reconfiguration of the pack is controlled by the BMS, these types of BMSs have been called a Reconfigurable BMS or R-BMS.

An R-BMS is capable of things like switching the configuration of the BMS between a series or parallel topology either globally or by stacks, isolating cells in case of fault or during equalization, regrouping cells according to aging parameters in different stacks, or adapting the configuration that best fits different services or types of loads. These systems incorporate a high number of switches and monitor circuitry, which can add to the BMS cost, and, in some cases, they will incorporate advanced algorithms that may need high computing power. Thus, several topologies have been proposed with different focuses, like increased performance of the system, high switching speed, less hardware complexity, or less computing requirements. Though this topic is an ongoing source of study, some topologies have recently been proposed, with several of these reviewed and evaluated in reference [175].

3.3.8. Battery Charging Profiles

A cell is considered discharged when its voltage falls below its cut-off voltage or when its SoC level falls below a value that is considered secure for its chemistry. At this point, the BMS will stop the discharge and get the battery ready to start the charging process. During charging, the BMS adjusts the supply of electrical energy controlling the current, voltage, and temperature until the battery voltage reaches a certain level or the current falls below the cut-off current. The charging profile, influenced by cell characteristics, may be calibrated by the BMS to adapt to the battery aging or differences between the cells within the battery pack.

Commonly, these schemes can be grouped under four major techniques:

-

Constant current charging (CC): Supplies a constant current until maximum voltage. This method is mainly used when charging NiCd, NiMH, and Li-ion batteries. Finding a satisfactory charging current value is challenging, as there is a compromise between the charging speed and temperature control (aging). A variation of the CC method is the “Pulse Charging” method (Figure 6a), where the current is supplied in pulses through the charging process [179].

-

Constant-Voltage charging (CV): This second method regulates the voltage supplied in parallel to a cell or a series of cells, making it constant through the charging process. It protects the battery from overvoltage, because it cannot supply higher voltage than the cell maximum voltage, and the charging current decreases gradually while the battery charges. A high current is necessary at the early stages of the charging process [180]. The main challenge for CV charging is selecting a proper voltage value that balances the charging speed and aging of the battery. By selecting a voltage that enables high currents between 15% and 80% of the SoC’s range, fast charging is possible [181,182], though it may accelerate the aging of the cells.

-

Constant-Current–Constant-Voltage charging (CC-CV): It combines the previous techniques, using CC in the early stages (bulk or bulking phase) until a safe threshold voltage is reached, and shifts to the CV method to complete the charging (absorption phase), as shown in Figure 6b. The charging time is mainly influenced by the calibration of the current used in the CC step, while the final SoC capacity is defined by the voltage selection during the CV step. This method is the most used for charging Li-ion and lead–acid cells. Variations include Constant-Power–Constant-Voltage (CP-CV) [183], shown in Figure 6c, and the Boost Charging [184], shown in Figure 6d. In the CP-CV method, the current and voltage are modified in such a way that constant power is supplied through the CC phase. In the Boost Charging method, the CC phase has two steps: first, large current value to elevate the voltage of the battery rapidly, and, second, a lower CC phase.

-

Multi-stage Constant-Current charging (MCC): The MCC method [185] is mainly used in fast charging by using different constant current values for different stages of the SoC of the battery (Figure 6e). This method is calibrated to use high currents at the beginning of the charge, which decrease as long as the SoC value increases, controlling the current value that passes through the battery and making profiles optimized to balance the speed of charge and battery aging. This method is considered suitable to charge lead–acid, NiMH, and Li-ion batteries. While cells subjected to the MCC may have shorter life cycles than cells subjected to the CC-CV method, the life cycle can be extended by a proper calibration of the charging steps and the use of adaptive methods [186,187]. A variation of the MCC method is the Variable Current Charging [188], where instead of steps, the current is modified linearly through the charging (Figure 6f).

In batteries with high self-discharge rates, e.g., lead–acid batteries, after finalizing the charge, the BMS will try to maintain a certain voltage, called the Float Voltage [189,190,191]. This voltage is calculated considering the self-discharge rate of the battery. Technically, any battery can be float charged, but it is uncommon for batteries with low self-discharge rates, such as lithium-ion batteries, as it may reduce their lifetime. A common tactic for floating lithium batteries is to apply this method only to batteries that need to maintain their charge during long-term storage to ensure energy availability.

When the battery is subjected to long-term storage, it may be kept at a storage voltage. A common technique in these cases consists of using a lower float charge during the storage operation. Because this level is not sufficient to compensate for self-discharge, the BMS periodically adjusts the voltage to maintain the SoC value.

3.4. Power Conditioning System (PCS) and Power Electronics

In a BESS, a Power Conditioning System (PCS) consists of devices that transform the power from the battery management system (BMS). This transformation suits the end application, such as a load or the grid. Typically, energy storage systems (ESSs) supply power at a DC voltage level, but when connecting to the grid, it must be converted to a higher AC voltage. This conversion improves the electrical source to align with the electrical network. These systems are sometimes referred to in academia as Power Electronics [192,193,194,195] and may include subsystems for thermal management and monitoring of the Power Conversion Units [192]. A general categorization of the PCS connection structures is presented in Figure 7.

3.4.1. Step-up Transformer Structures

In a centralized PCS configuration for a BESS, all battery packs connect to a common DC voltage bus (DC side), to which the PCS, acting as a DC/AC converter, is also connected (Figure 8a). Multiple battery clusters, which are combinations of packs with a shared BMS, may connect in parallel to the DC bus. Several PCS modules can also connect in parallel from the DC bus to a common AC bus, increasing system reliability or providing different ancillary services to the grid (Figure 8b).

The storage increase is achieved by connecting additional battery packs or clusters to the DC bus, and the power yield can be increased by connecting several centralized BESSs in parallel (Figure 8c). Additionally, multiple battery clusters and PCS systems may be connected to the DC bus, accommodating different applications or connecting to a common AC bus.

As shown in [196,197], a problem with this topology arises from differences in the current drawn from each parallel battery pack (DC side), causing disparities between packs throughout their aging cycle. Some packs may work in an overrated state, reducing the overall life of those packs and possibly affecting the whole BESS in a cascade effect.

Distributed PCS topologies address issues in the centralized BMS topology by connecting each battery pack to an individual PCS system, which may have one or two stages.

-

Distributed Single Stage DC/AC converter: (Figure 8d): Each battery pack has its own PCS unit that directly converts DC voltage to AC voltage, possibly integrated within the BMS. Although slightly less efficient, this topology allows individual power control per pack and the ability to handle different grid ancillary services in parallel. Other benefits include increased reliability and battery pack life cycle, as the packs are isolated.

-

Distributed Two Stages DC/DC + DC/AC converter: (Figure 8e): This is an improvement over the centralized topology, connecting each pack in parallel to a DC bus, and using a DC/DC converter to equalize the voltage between packs, thereby resolving differences in the drawable current.

-

Distributed Three Stages 2×DC/DC + DC/AC converter: (Figure 8f) A variation of the two stages topology for battery clusters connected to a larger DC/DC converter, with each battery pack equalized using a smaller voltage regulator.

An examination of the efficiency between the single and two stages topology under various grid connection scenarios is detailed in [198]. Additionally, several types of converters are available for both the DC and AC stages, with specific advantages depending on the grid application. A comprehensive description of these converter solutions can be found in [199,200].

3.4.2. No Step-up Transformer (Transformerless) Structures

These structures aim to eliminate the line-frequency transformer commonly used to connect the BESS solutions to the grid, reducing both the cost and space of the installation [193,201]. This can be achieved either with the Series Connection of Semiconductors (or Series Connection of Submodules), with the Cascaded H-Bridge Converters, or with the Modular Multilevel Converter.

The Series Connection of Semiconductors topology can employ several configurations, as described in [193,201]. However, due to its complex construction and the added cost of converters from the numerous switches used, this technology aims to achieve lower losses at low frequencies. This objective increases the cost due to the need for high-performance LCL filters. A detailed description of the inner workings of these topologies can be found in [202,203]. The Series Connection of Submodules topologies are more common among transformerless options, owing to a lower level of energy losses combined with higher modularity and scalability.

Cascade H-Bridge Converters consist of several single-phase H-bridge cells connected in each phase of the electrical grid, allowing for both star (Figure 9a) and delta (Figure 9b) configurations, depending on the grid characteristics. The star topology is typically more cost-effective, while the delta configuration is better suited for handling grid unbalance. These topologies usually involve a DC/AC converter in the form of an H-Bridge cell per battery pack without a DC/DC stage, although some authors suggest that adding a DC/DC stage could improve battery life expectancy [204].

Modular Multilevel Converters are commonly described in the literature as a double star connection (Figure 9c) with DC/AC converters shaped like chopper cells. Variants using bridge inverter cells are shown in [200], as this topology requires six branches instead of three, like the Cascade H-Bridge topology [205]. One advantage of this design is its flexibility in connecting energy storage elements, whether directly to the DC link, parallel to the double star branches as a large battery cluster, or distributed across the double star branches, similarly to the Cascade H-Bridge connection [193,205,206].

The major drawback of transformerless topologies is the absence of galvanic isolation. When isolation is required for safety and to minimize current leakage, an alternative is to use a solid-state transformer (SST). Although the SST reduces the size and weight of the magnetic core compared to a conventional transformer, it currently presents some disadvantages, such as increased cost, reduced reliability, slightly lower efficiency, and challenges for grid integration [207]. Variations of transformerless topologies incorporating a SST are discussed in [193].

3.5. Challenges for Large-Scale Stationary Applications

The large deployment of BESSs expected in the coming years and decades, driven by the growing penetration of renewables, poses significant challenges that need to be addressed in the design of the BMS:

-

Efficiency: Minimizing energy losses while extending battery life is crucial, especially in large-scale applications, where even the smallest energy loss can accumulate to MWh or GWh over the BESS life cycle.

-

Electrical and thermal security: In large systems, monitoring the electrical and thermal parameters of the battery ensures a long life cycle, on-site security, and appropriate energy delivery to the grid. The BMS must incorporate functions that monitor the operation of the batteries according to legislation while coordinating with the BTMS, PMS, and other systems to ensure security.

-

Data security: Because BESSs are frequently connected to the main grid or integrated into microgrids, ensuring secure communication between the BMS and other systems is of utmost importance to prevent data breaches. In this context, the IEC 62351 standard identifies various risks and vulnerabilities. For a comprehensive understanding of the cyberphysical security challenges in BESSs, including communication and hardware vulnerabilities, refer to the review presented in [208].

-

Runtime: Profitability is closely linked to runtime hours, underscoring the importance of minimizing the rate of battery aging while meeting the electrical grid or load requirements. For example, in systems where high energy availability and short response times are crucial, optimizing the charge and discharge profiles in the BMS becomes imperative.

-

Standardization: With the increasing trend of incorporating smart sensing technology inside individual cells and battery packs, there arises a necessity for standardization in communication mechanisms between these sensors (intra-battery communications) and the BMS. Standardization in this area is crucial for enhancing the overall system efficiency.

-

Repurpose of batteries: The future adoption of electric cars and other vehicles creates a new need for repurposing EOL batteries for these applications. Generally, a battery is considered in an EOL state when its capacity drops below 80% of its specified nominal value. Mobility use cases must balance battery capacity with weight and volume, while stationary applications may have more flexibility. As a result, these batteries find a new use as a BESS, which requires the development of new algorithms for battery equalization from different manufacturers, as well as tailored aging and SoC algorithms for these batteries.

3.6. Promising Technologies and Research

Several upcoming technologies have the potential to significantly change the way the BMS and BESS work and interact with each other:

-

Self-healing: Self-healing batteries with polymer electrolytes materials have been introduced as a topic of research and an objective of the Battery 2030+ initiative, promoted by the European Union. Self-healing materials seek to fix chemical and physical degradation processes on the batteries, such as electrode cracking, loss of electrical connectivity, electrolyte degradation, and lithium plating. New necessities will follow in BMS design, such as readaptation of the aging algorithms for tracking the functional status of the cells [209]. Through the integration of smart sensors on the battery cells, the degradation process can be tracked, introducing new requirements for the BMS, such as interacting with these sensors and implementing Big Data models (BIG-BMS) [210]. When early aging is detected, the BIG-BMS would emit a signal to activate the polymeric actuators, initiating the self-healing process. Inbuilt cell sensing is highly promoted in the Battery 2030+ initiative, with various projects, such as SENSIBAT [211], INSTABAT [212], and SPARTACUS [213].

-

Digital Twins and cloud processing: Digital Twins are mathematical models that provide accurate representations of real systems, and they are a promising approach in BMS design [36,214]. By incorporating this technique in a BESS, a model of the charging and discharging process, as well as the BMS work cycle, can be generated. By comparing the model with measured data and analyzing deviations between the model and real data, functional errors and malfunctions can be detected, allowing for adjustments of BMS parameters to mitigate battery aging and enhance the safety and reliability of the system. Furthermore, the data can be repurposed to design systems that adapt more accurately to real-life conditions in the future.

• Because processing such data can be hardware demanding, especially in large systems, it is proposed to integrate these models in cloud-based BMS devices [37,215,216]. A cloud-based BMS ultimately follows a distributed BMS topology, where the on-site BMSs just incorporate the main functions for security but are subject to orders and corrections from the offsite main BMS.
• Another approach that complements the cloud model is the blockchain architecture. In a blockchain network, all BMS devices are connected in a peer-to-peer fashion, forming a distributed network where each BMS acts as a node exchanging data. These approaches can incorporate typical blockchain technologies, such as the consensus mechanism and cryptographic encryption, to enhance the security of the data transfer. Some blockchain approaches in a BESS are meant to implement scheduling techniques for battery charging [217], firmware deployment [218], and enhanced cybersecurity [208].

-

Field Programmable Gate Array (FPGA): An FPGA is a semiconductor device based around a matrix of configurable logic blocks, enabling it to be reprogrammed for a wide array of applications at the hardware level, resulting in faster and more accurate response times than software solutions. FPGAs stand out due to their reconfigurability, distinguishing them from Application Specific Integrated Circuits (ASICs), which are custom manufactured for specific design tasks. FPGAs have been recommended for designing advanced BMS devices. They function as hardware emulators capable of replicating specific hardware, making them ideal for running Hardware-in-the-Loop simulations of BMS devices [219,220]. Additionally, they have been proposed as accelerators for faster SoC, SoH, and SOA calculations, and as adaptable BMS design solutions [221,222,223].

4. Techno-Economic Assessment

The utilization of a BESS to provide diverse grid services at various voltage levels is experiencing a significant growth. In all of these applications, the cost of the BESS becomes a critical aspect that demands careful consideration and analysis [224]. The increasingly prominent penetration of renewable energies and the progressive decrease in BESS costs will lead to batteries being increasingly implemented to provide all these services. Battery storage investment soared in 2022, doubling its previous levels. This surge was fueled by increased interest and financial support from institutional investors and renewable energy developers [225].

The following sections provide, based on the reviewed literature, a detailed overview of the benefits and services that BESSs can provide at different levels of the electric grid (Section 4.1), the costs associated with their installation and operation (Section 4.2), and the key economic indicators used to assess battery storage projects (Section 4.3).

4.1. Benefits and Services Provided by BESSs

The following are descriptions of the various services that a battery energy storage system (BESS) can offer at different levels of the electric grid. Additionally, references to the literature are provided where technical and economic aspects of these services are addressed.

• Self-consumption [226,227,228,229,230,231,232]: A BESS can improve the self-consumption of generated electricity by storing excess energy during periods of low demand and supplying it during peak consumption times. This allows consumers to reduce their reliance on the grid and optimize their energy usage, resulting in cost savings and increased energy efficiency. Numerous articles analyze, from a technical and economic perspective, the performance of self-consumption installations with BESSs.
• Peak shaving [233,234,235,236,237]: Battery energy storage systems can help mitigate peak demand by discharging stored energy during periods of high electricity consumption. By reducing the peak load on the grid, batteries can help avoid the need for costly infrastructure upgrades and alleviate strain on the power system during peak periods, leading to improved grid stability and reliability.
• Load Shifting [238,239,240,241,242]: Load shifting refers to the practice of adjusting the timing of energy consumption to take advantage of more favorable conditions, such as lower electricity prices or increased availability of renewable energy.
• Energy Arbitrage [243,244,245,246,247]: This is the practice of purchasing electricity from the grid during periods when electricity prices are low and storing it in the BESS and then selling it back to the grid when electricity prices are higher.
• Ancillary services [248,249,250,251,252,253,254,255,256,257]: BESSs can provide ancillary services, such as frequency regulation and voltage support, amongst others, to assist in grid stability and reliability. They can rapidly respond to fluctuations in electricity supply and demand, injecting or absorbing power as needed to help maintain the balance between generation and consumption, thus supporting the overall grid operation.
• Curtailment minimization [252,254,255,258,259,260]: Curtailment minimization is a service provided by a BESS to mitigate or prevent the curtailment of renewable energy generation. In situations where the renewable energy supply exceeds the demand or grid capacity, a BESS can absorb the excess energy and store it for later use. By avoiding curtailment, valuable renewable energy resources are preserved and can be utilized effectively.
• Distribution grid upgrade deferral [250,252,261,262,263]: Batteries deployed at the distribution level can defer or avoid the need for costly upgrades to distribution infrastructure. By storing and releasing energy as required, batteries can help manage local demand and supply imbalances, reduce grid congestion, and improve the overall efficiency and reliability of the distribution grid.
• Transmission grid upgrade deferral [264,265,266]: Similarly, batteries located at the transmission level can defer or eliminate the need for costly transmission grid upgrades. By providing grid support services and dynamic power flow control, batteries can optimize the utilization of existing transmission infrastructure and enhance the capacity and reliability of the transmission grid.
• Backup power [250,267,268,269,270]: A BESS can serve as a backup power source during grid outages or emergencies. When the grid fails, the stored energy in the batteries can be used to power critical loads, providing uninterrupted electricity supply and enhancing grid resilience.
• Energy islands [271,272,273,274]: BESSs are an important component of energy islands, which are self-contained energy systems that operate independently of the main grid. Energy islands can integrate various renewable energy sources and BESSs.

Services, such as arbitrage, ancillary services, and curtailment minimization, will gain importance due to the greater integration of renewables into the grid. Others, like load shifting, peak shaving, and grid upgrade deferral, will become more significant in line with the ongoing trend of electrification across diverse sectors of the economy, such as heating and transportation. Additionally, emerging business models that benefit from BESSs, such as energy storage as a service (ESaaS), long-term storage, virtual power plants, and microgrids, will play a role in shaping the energy landscape in the coming years.

4.2. Cost Breakdown of a BESS

The cost breakdown of a BESS can be divided into two primary categories: capital expenditures (CAPEX) and operating expenses (OPEX). A detailed overview, obtained from the literature [275,276,277], of the costs included in each of these terms is provided bellow.

4.2.1. Capital Expenditures (CAPEX)

-

Battery Cost: This includes the cost of the batteries themselves, which are the primary component of a BESS. Battery costs can vary depending on the chemistry, capacity, and manufacturer.

-

Power Conversion System: A BESS requires power conversion equipment, such as inverters, transformers, and switchgear, to convert DC power from the batteries into AC power for use in the electrical grid.

-

Installation and Construction: This includes the cost of site preparation, installation of batteries and power conversion equipment, and other construction-related expenses.

-

Control and Monitoring Systems (BMS): A BESS requires sophisticated control and monitoring systems to ensure efficient operation and safety.

-

Electrical Interconnection: A BESS needs to be connected to the electrical grid, and the cost of interconnection equipment, such as transformers and switchgear, is included in the CAPEX.

-

Other Ancillary Equipment: This includes additional equipment, like cooling systems, fire suppression systems, and safety measures required for the BESS installation, amongst others.

-

Other costs: Development cost, environmental studies and permitting, legal fees, etc.

4.2.2. Operating Expenses (OPEX)

-

Maintenance and Repairs: A BESS requires regular maintenance and occasional repairs to ensure optimal performance. This includes inspections, battery replacements, and upkeep of power conversion equipment.

-

Operations and Monitoring: Ongoing operational costs include personnel salaries for monitoring and controlling the system, as well as any associated software or data management costs.

-

Insurance and Permits: BESS installations may require insurance coverage against potential risks, and permits may be necessary for construction and operation.

-

Grid Connection Fees: Some jurisdictions impose fees for connecting to the electrical grid or utilizing grid services, which contribute to the OPEX.

-

Decommissioning and Disposal: Eventually, when a BESS reaches the end of its useful life, there will be costs associated with decommissioning and environmentally responsible disposal of the equipment.

4.3. Financial Metrics

Evaluating the economic viability of BESS projects requires the use of financial metrics or indicators. These metrics provide insights into the profitability, costs, and financial feasibility of such projects. By considering various factors, such as investment costs, operational expenses, revenue streams, and the time value of money, these indicators help assess the long-term financial performance and potential returns on investment for BESS projects. Understanding and analyzing these economic indicators is essential for making informed decisions in the planning and implementation of BESS projects.

Table 9. Financial metrics and references to specific studies (2018–2023) that have utilized them.

Financial Metrics	Definition	Formula	Literature Where It Is Utilized 1
Net Present Value (NPV)	NPV measures the profitability of an investment by calculating the difference between the present value of cash inflows and outflows.	$NPV=\sum _{t=1}^n\frac{CF\left(t\right)}{{\left(1+r\right)}^t}-{CF}_0$ CF(t): expected cash flow year t CF0: initial investment r: discount rate.	[272,278,279,280,281,282,283,284,285,286,287,288,289,290,291,292,293,294,295,296,297,298,299,300,301,302,303,304,305,306,307]
Levelized Costs of Storage (LCOS)	It is a metric used in the energy sector, specifically for battery projects, representing the average cost of deploying and operating an energy storage system over its lifetime.	$LCOS=\sum _{t=0}^n\frac{cost\left(t\right)}{{\left(1+r\right)}^t}/\sum _{t=1}^n\frac{E_{out}\left(t\right)}{{\left(1+r\right)}^t}$ cost(t): Cost in year t Eout(t): Electricity discharged year t r: discount rate.	[290,308,309,310,311,312,313,314,315,316,317,318,319]
Levelized Costs of Energy (LCOE)	It is a measure of the average net present cost of electricity generation for a generator (in this case, generator + storage) over its lifetime.	$LCOE=\sum _{t=0}^n\frac{cost\left(t\right)}{{\left(1+r\right)}^t}/\sum _{t=1}^n\frac{E\left(t\right)}{{\left(1+r\right)}^t}$ cost(t): Cost in year t E(t): Electricity generated year t r: discount rate.	[272,284,301,302,305,306,308,310,311,312,314,318,320,321,322,323,324,325,326,327,328,329,330,331,332,333,334,335]
Internal Rate of Return (IRR)	IRR is the discount rate that equates the net present value of cash inflows and outflows to zero.	$0=\sum _{t=1}^n\frac{CFt}{{\left(1+IRR\right)}^t}-{CF}_0$ CFt: expected cash flow in year t CF0: initial investment r: discount rate.	[257,285,288,289,290,295,298,299,305,328,336,337,338,339,340]
Net Present Cost (NPC)	NPC evaluates the total cost of an investment over its entire life cycle, including initial investment and operational and maintenance costs.	$NPC=\sum _{t=0}^n\frac{cost\left(t\right)}{{\left(1+r\right)}^t}$t cost(t): Cost in year t r: discount rate.	[302,324,326,328,330,333,334,341,342,343]
Payback Period (PP)	The payback period measures the time required for the cumulative cash inflows from an investment to equal the initial investment cost.	$PP=t-1+\frac{Ut}{CFt}$ t: Break-even year Ut: Unrecovered amount in year t CFt: expected cash flow in year t.	[238,278,279,289,295,299,300,305,306,325,338,344,345,346,347,348,349,350]
Discounted Payback (DPB)	Similar to payback period, discounted payback considers the time value of money by discounting the cash flows.	$PP=t_d-1+\frac{U{t}_d}{CF{t}_d}$ td: Break-even year (discounted) Utd: Unrecovered amount in year t (discounted) CFtd: expected cash flow in year t (discounted).	[238,286,289,295,298,303,328,339,344,347,348,351]
Equivalent Annual Cost (EAC)	EAC represents the average annual cost of an investment, considering both initial and recurring costs over its expected lifespan.	$EAC=\sum _{t=1}^n\frac{cost\left(t\right)}{{\left(1+r\right)}^t}-{cost}_0$ cost(t): expected cost in year t cost0: initial investment r: discount rate.	[326,329,352,353,354,355,356,357,358]

¹ Specific studies, research papers, or publications that have utilized the mentioned financial metrics in the evaluation of battery energy storage systems projects.

shows the definitions and formulas of financial metrics used in BESS projects as well as references to specific studies from 2018 to 2023 that have utilized the mentioned financial metrics.

The analyzed papers examine a wide variety of projects where batteries provide different services in diverse contexts, leading to a significant variability in financial indicators. Assessing the economic feasibility of projects involving battery energy storage systems presents a challenging task [224]. The measurement of economic and financial benefits derived from a BESS covers a wide range of complexities, varying from relatively simple scenarios where the primary effect is on quantity to more challenging situations where the focus is on quality [276]. This level of difficulty is also closely linked to the time scale involved. For instance, when a BESS operates over several hours (e.g., shifting PV output from mid-day to the evening peak), the underlying principle is straightforward in theory, but the practical estimation of economic benefits can become challenging due to modeling and data requirements. However, estimating the economic benefits becomes even more arduous when the benefits occur at time scales of seconds or less, as rapid effects require more intricate analysis [276].

The literature indicates that the economic evaluation of a BESS is significantly influenced by various factors, including the range of services provided by the BESS, the prevailing electricity prices, the availability of alternative flexibility options and markets, the storage operating time, the characteristics of the overall system, sector coupling, government policies, incentives, the geographical location, the grid infrastructure, or energy demand patterns, among others. Haas et al. [359] indicate that storage capacities should be introduced only when it is evident that the generation of electricity from variable renewables will also grow, resulting in an anticipated surplus of electricity.

Some of the analyzed papers view BESSs as unattractive investments, although sometimes they note that some of the indirect benefits, like outage reduction, enhanced power quality, deferred distribution investment, or reduction of peaking plant capacity, are not analyzed. Mostafa et al. [329] examined the costs of various long-term, medium-term, and short-term energy storage technologies, considering different power and energy ratings. They determine that pumped hydro storage is the most competitive option for long-term storage, while sodium–sulfur batteries excel in the medium term, and supercapacitors are the top choice for short-term storage technologies [359]. Topalović et al. [360] assessed pumped hydro storage and battery storage technologies in the Western Balkans region. Their findings indicate that pumped hydro storage remains the most cost-effective storage technology for large-scale energy storage in this area. Echevarria Camarero et al. [231] analyzed the advantages of incorporating batteries into PV self-consumption systems, considering various self-consumption models, several industrial consumer profiles, and different locations across Spain. The research findings indicate that, considering current electricity prices, utilizing batteries is not a profitable option in the cases studied.

On the other hand, studies highlight that a BESS can be profitable in certain contexts. Various research papers have demonstrated profitability in different applications, including participation in balancing markets, fast frequency regulation, and peer-to-peer trading. However, profitability depends on certain factors, such as market conditions, the technology, and the specific application of the BESS. Laslett [361] found that utilizing high levels of distributed renewable energy and relying solely on batteries as the primary storage technology can result in a cost-effective solution for meeting the energy demands of a large-scale electrical grid in Australia. Canevese and Gatti [362] found that a BESS can participate in the Italian Balancing Market and generate profits, depending on the optimization approach used and the penalty coefficient related to battery cycling aging. Hu et al. [363] analyzed the potential utilization of BESSs in major European electricity markets and found that a BESS is not feasible for energy arbitrage in most markets but has significant potential in providing frequency support services. According to Schneider et al. [364], economically viable options for deploying rechargeable batteries in Switzerland for demand peak shaving and price arbitrage business can be achieved through the prudent selection of energy capacity and power unit sizes. In their study, Grzanic et al. [365] examine the profitability of investing in PV and battery energy storage integrated with EV charging stations in Croatia, and the research reveals that such investments can be profitable, especially considering the current retail market options.

An undeniable fact is that, as previously mentioned, battery energy storage systems are witnessing a remarkable surge in usage, offering diverse grid services across various voltage levels. Notably, there has been substantial growth in battery storage investment in recent years, supported by initiatives such as the US Inflation Reduction Act and incentives in Europe, Australia, China, Japan, and Korea, amongst others [225]. Looking ahead, the increasing investment in battery storage has great potential because of favorable economics, the integration of renewable energy, and policy support.

5. Conclusions

The current energy context, with high penetration of renewable energy sources of an intermittent nature, will require a large deployment of energy storage, among other aspects, to catalyze a real and effective transition towards a green economy. Batteries (electrochemical storage) are among the alternatives for energy storage, and there are very different chemistries available on the market. This review has tried to evaluate the subjects and advances in modeling of these batteries and their BMSs from a very generic perspective, i.e., as independently as possible from the technology chosen. The main conclusions and future lines of research are described in the following paragraphs.

Regarding battery modeling, equivalent circuit models (ECMs) are the preferred option to use in battery management systems due to their relative simplicity and high accuracy (error in order of mV). This is true for various battery technologies, as the nature of the main electrochemical phenomena is fundamentally the same. Various factors, such as the temperature, state of charge (SoC), and current rate, significantly influence battery parameters, and the phenomenon of hysteresis often goes unaddressed. Enhanced hysteresis models could be developed considering these factors. The fitting of parameters must accurately reflect the characteristics of the application context, and the growing popularity of data-driven models, thanks to abundant data, leads to notably precise results.

Concerning the SoC, there is a generally accepted definition, but there is not a strict universal consensus regarding the specifics, such as the capacity used. Broadly speaking, the most common methods found in a BMS are state-observer models, particularly Kalman-Filter-based estimation methods, as they provide accurate, error-bounded, and robust SoC estimations if tuned properly. Data-driven methods are also used effectively to obtain SoC estimations and are often used in combination with state-observer models.

SoH suffers from the same definition variability, as there are multiple definitions available. Battery aging characterization may benefit from having different SoH indicators regarding different aspects such as capacity and impedance, with proper naming to avoid ambiguity. Model-based estimation methods and state-observer models have proven to be adequate methods to estimate the SoH of batteries, but it is also important to consider that different battery technologies present particularities that can be used to correlate with SoH indicators. Again, data-driven methods popularity has increased significantly in estimating state of health, presenting remarkable results.

Regarding the BMS, multiple factors need to considered for its design, tailored to the characteristics of the battery and the end application of the BESS. Battery pack configuration, speed of charge, and maximum discharge capacity are all factors to be considered, along with the safety of operation and efficiency of the system.

For major deployment in large stationary applications, it will be necessary to reconcile the speed of charge and power delivery for grid applications with the aging of the battery cells. In this regard, meticulously defined charging profiles for each grid application may be needed. Combining this with R-BMS, intra-cell measuring, and self-healing materials presents potential for increasing the profitability of BESSs. The introduction of second life batteries from the EV sector may also influence the speed of deployment of BESSs.

In large storage applications, thermal, electrical, and data security are also important criteria to take into account by the BMS. Digital twin technology can be a good solution to detect early failures in the batteries, while the implementation of blockchain technology can increase data security, although its feasibility will depend on the regulatory framework of each country.

To conclude, this review has delved into the techno-economic perspective of BESSs. It is important to emphasize the extensive array of services and benefits that batteries can provide at various voltage levels within the power grid. The analysis of the existing literature reveals both positive and negative economic outcomes depending on numerous factors but leads to a strong conclusion that the investment attractiveness of BESSs will undoubtedly increase as the energy transition progresses.