1. Introduction
Due to the serious environmental pollution problems caused by the exhaust emissions emitted by diesel and gasoline vehicles, electric vehicles (EVs) have attracted more and more attention in recent years and have become a new research focus [
1]. The researchers in [
2] provide an efficient strategy for the optimal scheduling of large-scale EVs connected to a predictable wind power grid. In contrast, the optimal scheduling of a predictable wind power grid provides an effective solution. In [
3], a novel, innovative environmental mathematical formulation and corresponding deep reinforcement learning algorithm applied to EVs are proposed to improve the self-consumption of photovoltaic power generation and the charging state of EVs during charging. It is proved through experiments that it can maximize the
SOCial benefits of the transportation system and is significant to the sustainable development of urban transportation [
4]. Nowadays, EVs are powered by a battery pack consisting of multiple battery units, and users can access real-time information about each battery, such as state-of-charge (
SOC) and state-of-health (
SOH), by the BMS [
5]. These parameters are the key performance indicators of each battery unit in the EV battery pack.
SOC consistency refers to the degree of consistency of the state of charge of each battery in a battery pack consisting of multiple batteries. Due to the manufacturing process of battery cells, age, charge and discharge times, environmental conditions, and other factors, the capacity and performance of each battery cell will gradually appear different, resulting in some battery cells in the battery pack than other battery cells earlier than the entire state of charge or discharge state. Therefore, it is necessary to ensure the consistency of the
SOC of each cell in the battery pack to improve the efficiency and reliability of the battery pack. Researchers in [
6] have proposed an active balancing method based on two flyback converters for series-connected battery packs. The balancing energy can be transferred between the whole battery and any individual cell. In [
7], authors propose a novel lossless two-stage equalization circuit topology based on a conventional buck-boost circuit to achieve equalization of series-connected lithium-ion battery packs with higher efficiency and lower cost—the proposed topology efficiently consensus lithium-ion battery packs without adding additional devices. The experiment in [
8] showed that 24 lithium-iron-phosphate battery packs composed of series-connected batteries were tested for cycling performance. After 300 cycles, significant differences were shown in the
SOC level of individual batteries, and some batteries significantly deviated from the average
SOC, exhibiting severe “sweep broom” effects. To prevent over-discharging of low
SOC batteries, the battery pack must be prematurely terminated, resulting in a decrease in the actual usable capacity of the battery packs. Furthermore, based on the data from China’s Zhangbei Reserve Power Station, the station’s battery packs consist of lithium-ion batteries of the same batch and type, divided into six battery packs, each of which has been in operation for two years [
9]. The issue of varying battery capacities within the group is becoming more pronounced. In one group, the battery with the highest
SOH had its capacity reduced by
, while the battery with the lowest
SOH had its capacity reduced by
. The inconsistencies in
SOH lead to significantly decreased battery capacity, performances, and lifespans, which indirectly impact the consistency of
SOC. As a result, a consensus of
SOC and
SOH must be considered simultaneously.
The first problem to be solved is how to accurately estimate
SOC and
SOH. In [
10], authors constructed a distributed control-oriented electro-thermal coupled model and used an improved parameter identification method to construct an offline model parameter database. Recursive least squares and particle swarm optimization methods with variable forgetting factors were used to identify the electrical and thermal parameters. A
SOC-corrected core temperature estimation technique finally moves forward. In [
11], the authors proposed a fully distributed state estimation method for power systems based on weighted least squares and graph theory. In addition, unlike the existing methods, the method proposed in this paper is a bus-level DSE method, which does not require the power system to be divided into multiple regions. The fully distributed approach of the method is enlightening for
SOC estimation of multiple batteries. In [
12], the researchers employ the open-circuit voltage (OCV) to estimate
SOC accurately. In [
13], the authors utilize a combination of BP neural networks and PSO algorithms for
SOC estimation. In [
14], a novel
SOC closed-loop estimation algorithm is introduced, based upon the fusion of the ESP model and ampere-hour integration, which combines the merits of both methods and enhances the accuracy of
SOC estimation. Regarding
SOH estimation, the study by [
15] proposes a method based upon an improved ant lion optimization algorithm and support vector regression (IALO-SVR), providing more precise estimations than conventional methods. In [
16], the author suggests using an extended Kalman filter-recursive least squares parameter identification method based on the second-order RC equivalent circuit model. In [
17], a new deep learning network framework is proposed to achieve
SOH estimation. To improve the accuracy of the estimates, the researchers in [
18] proposed a method based on a hybrid neural network called Gate Recursive Unit-Convolutional Neural Network, which can learn the shared information and time dependence of charging profiles through deep learning techniques, and can utilize the newly observed charging profiles (e.g., voltages, currents, and temperatures) to estimate
SOH. In [
19], proposed as a means of estimating
SOC and
SOH, an adaptive segmented equivalent circuit model based on the extended Kalman filter is illustrated in the degradation study and the model validation procedure. The findings demonstrate the applicability of the segmented parameter adaptation described in this study to various battery chemistries and aging conditions. In [
20], the authors propose an improved Remora optimization algorithm. After detecting the optimal values of the parameters, the
SOC will be evaluated by a dual adaptive Kalman filtering algorithm. Then, the
SOH is estimated based on the predicted cell
SOC. In [
21], in order to improve the accuracy of real-time estimation of battery state, a novel back-propagation neural network-double extended Kalman filter based on finite-element memory recursive least squares is proposed for the synergistic estimation of
SOC and
SOH in lithium-ion batteries by creating a second-order equivalent circuit model. The researchers in [
22] proposed an adaptive double square root Kalman filter with a resting region to accomplish the estimation while accounting for aging effects. The first filter estimates the
SOC and
SOH, and the second adaptively updates the drift model parameters based on the dormant region. However, these algorithms have two drawbacks: they can only estimate either
SOC or
SOH, or they estimate both but need to be simplified.
Then, the consensus between
SOC and
SOH should be considered. In [
23], the authors compare the
SOC consensus problems under continuous systems, discrete systems, self-triggering, and event-triggering mechanisms (ETM). In [
24], a distributed model predictive control approach is proposed to achieve dynamic response to load perturbations and maintain
SOC consistency for heterogeneous energy storage systems. The authors suggest an adaptive power management plan for standalone microgrids to address these issues based on the battery’s available
SOC. Distributed energy resources can be smoothly transitioned from maximum power point tracking mode to adaptive sag mode and vice versa, using the suggested battery
SOC-based adaptive power management sag control method [
25]. In [
26], this article proposes a further multi-layer
SOH consensus scheme to consensus
SOH across all batteries through a comprehensive portfolio of
SOH’s large-scale BESS balancing strategy and commercial battery balancing technology. In [
27], this paper presents a battery management approach for monitoring and controlling temperature,
SOC,
SOH, etc., enhancing the efficiency of rechargeable batteries. This approach provides some insights into our work. However, these algorithms can only perform observation processing on one parameter at a time, which cannot meet the multi-parameter situation in the actual situation. Therefore, it needs to be improved.
It is well known that network systems are always limited by energy and bandwidth, which is why the ETM was introduced. In [
28,
29], the authors proposed an ETM with fixed thresholds. In [
30], the authors considered the
SOC consensus under the ETM and added the power supply and demand consensus in the context of energy storage systems. In [
31,
32], two different ETMs were proposed, and satisfactory results were achieved. However, these ETMs usually adopt fixed thresholds, which do not consider the change in information during operation and cause the waste of this information. Based on these works, this study aims to propose a new design for the ETM. During the information-gathering phase, the trigger condition is established by the value of the battery itself. In the information transmission stage, the trigger condition is determined by the transmitted information and the collected information, thereby enhancing the flexibility of the overall mechanism and eliminating the drawback of information waste caused by fixed thresholds of the existing mechanism.
The novelties of this work are summarized as follows:
- (1)
To achieve accurate estimation of SOC and SOH of EV batteries, a novel distributed optimal Kalman consensus filter is proposed for a battery management system; it reduces the estimated error.
- (2)
A new event-triggered approach based on dynamic information is introduced to use information from the sensor and its neighbors entirely to save communication resources.
- (3)
To eliminate the impact of SOH on SOC, state estimation and consensus control are performed on both SOC and SOH simultaneously. This joint-consensus concept is proposed as more conducive to realistic battery management.
The remaining parts of this article are as follows: In
Section 2, the ET-DOKCF and two different ETMs are introduced.
Section 3 introduces some definitions of the ET-DOKCF, and a brief analysis of the system’s stability is provided.
Section 4 provides a filtering comparison of the three algorithms, a comparison of ETMs, anti-interference performances of the proposed algorithm under different real noises, and the display of consistency results.
Section 5 concludes this paper, and the derivation of ET-DOKCF is given in the
Appendix A.
3. Event-Triggered Distributed Optimal Kalman Consensus Filter and Balance Algorithm
For a smooth derivation, we first need some definitions. The priori estimated error and the posterior estimated error for the battery at time instant
k as:
Moreover, denoting the errors related to the event-triggered schedules as:
Then, the as
SOCiated error covariance matrices can be given by:
For convenience, we identify the notations that can define the matrices , , , , , by , , , , , , respectively.
Substituting (12) and (13) to (11), we have:
and similarly for sensor
j:
Based on the above definition, the ET-DOKCF algorithm is formulated as (20)–(24):
where
The above completes the accurate estimates of
SOC and
SOH. After obtaining the exact estimates of
SOC and
SOH for each cell, equalization is required in case of inconsistency. The algorithm originates from the study of multi-agent consistency, is based on algebraic graph theory, and can be realized using concise formulas. Let the number of batteries in the BMS be
M, and the estimated values of
SOC and
SOH for
ith battery are
and
. Denote the
and
, then the balance algorithm is:
where
I is the unit matrix, and
and
are constant parameters,
T is the sampling period, and
L is the Laplace matrix for the topological diagram between batteries. This balance algorithm possesses the distinct advantage that it enables the
SOC and
SOH of the system to be consistent at all times in a simple yet effective manner, thus avoiding the need to introduce cumbersome and complex steps and formulas to achieve this goal. In addition, the algorithm requires a relatively small number of parameters and formulas, further simplifying the implementation process. With this feasible approach, the consistency of
SOC and
SOH can be efficiently maintained without burdening the implementer.
The above is a complete algorithm for estimating and balancing battery SOC and SOH based on ET-DOKCF, and the whole process can be described by Algorithm 1 as follows:
Algorithm 1 Estimation and Balance of SOC and SOH Based on ET-DOKCF |
- 1:
Provide the initial values , and at for each sensor, iteration step K - 2:
while do - 3:
while do - 4:
Obtain measurement - 5:
if trigger condition in (6) = True then and - 6:
end if - 7:
if trigger condition in (7) = True then and - 8:
end if - 9:
Compute - 10:
Compute matrices in (14), (15), (16), and (17) - 11:
Compute and with results above - 12:
Compute the optimal KCF based on ETM - 13:
Update the state of the local information filter and - 14:
end while - 15:
end while - 16:
Balance SOC and SOH for each battery with the results above by (25)
|
Remark 1. The research on can be categorized into numerical and matrix forms. Among numerous findings, selecting the appropriate to maintain system stability without compromising algorithm efficiency remains a pending issue. In the context of this article, with prolonged usage and frequent charging and discharging, the BMS performance degrades, and unstable factors arise in the presence of complex computing systems. Consequently, we adopt the former approach where is assumed as a small constant, such as 0.1. The value can be adjusted according to usage scenarios, ensuring system stability while reducing computational complexity. Furthermore, the ETMs in this paper eliminate the possibility of the Zeno phenomenon based on discrete time.