Energy Management Strategy for Distributed Photovoltaic 5G Base Station DC Microgrid Integrated with the CF-P&O-INC MPPT Algorithm

Abstract

With its technical advantages of high speed, low latency, and broad connectivity, fifth-generation mobile communication technology has brought about unprecedented development in numerous vertical application scenarios. However, the high energy consumption and expansion difficulties of 5G infrastructure have become the main obstacles restricting its widespread application. Therefore, aiming to optimize the energy utilization efficiency of 5G base stations, a novel distributed photovoltaic 5G base station DC microgrid structure and an energy management strategy based on the Curve Fitting–Perturb and Observe–Incremental Conductance (CF-P&O-INC) Maximum Power Point Tracking (MPPT) algorithm from the perspectives of energy and information flows are proposed. Simulation results show that the proposed MPPT algorithm can increase the efficiency to 99.95% and 99.82% under uniform irradiation and partial shading, respectively. Under the proposed strategy, when the base station load changes drastically, the voltage fluctuation of the DC bus is less than 1.875%, and returns to a steady state within 0.07s, alleviating the high energy consumption of 5G base stations effectively and achieving coordinated optimization management of various types of energy in multi-source power supply systems.

1. Introduction

With the advancement of information and communication technologies (ICT), fifth-generation mobile communication technology (5G) offers high-bandwidth, high-capacity, and low-latency communication [1], exerting significant economic benefits and profound social impacts on the development of the digital economy [2], smart cities [3], and intelligent manufacturing [4]. Currently, 5G network coverage is continuously expanding, while the energy consumption of 5G base stations is rapidly increasing [5]. Through testing and verification in trial commercial networks, the power consumption of a single 5G base station is estimated to be around 3.5–3.9 kW, which is 3–4 times that of 4G base stations [6]. Additionally, under the same coverage area, 5G networks require more antennas and denser base stations. Overall, the comprehensive power consumption of 5G base stations is expected to be 9–10 times that of 4G base stations. The sharp increase in energy consumption imposes enormous pressure on grid power supply and operation costs [7], thus attracting increasing attention regarding the feasibility of photovoltaic-powered 5G base stations as a solution to alleviate the high power consumption, expansion difficulties, and operational costs.

In the field of research on photovoltaic-powered 5G base stations, a commonly encountered structure is to directly connect the photovoltaic (PV) array in a series-parallel configuration through high-power MPPT controllers and direct current/alternating current (DC/AC) converters to the AC bus, and then rectify it to supply power to the 48 V DC bus [8]. This installation method is simple and integrates photovoltaics into the power supply system of 5G base stations. However, this structure involves multiple stages, and after photovoltaic energy is integrated, it needs to undergo multiple power conversions, leading to a decrease in overall energy conversion efficiency. Moreover, it results in an excessive number of power electronic converters in the system, thereby increasing deployment costs [9].

On the other hand, in photovoltaic power generation systems, the efficiency of electricity generation is low due to the influence of external environmental factors such as solar irradiance, temperature, and load, as well as internal photovoltaic characteristics. Compared with the suburbs, the deployment of base stations in the city is more intensive [10]. However, in the city, limited by the geographical location and building shading, the deployment of PV cells near the base station is more fragmented, and the tilt angle and shading conditions are different [11], In traditional series-parallel photovoltaic arrays, the decline in power output of individual photovoltaic cells under partial shading conditions can impact other photovoltaic cells, thereby significantly reducing overall electricity generation efficiency. At the same time, many of the existing PV MPPT technologies are not applicable to the partially shaded environment caused by cloudy conditions and other factors [12]. In addition, despite the potential advantages of PV energy access to base stations, the intermittent and fluctuating nature of renewable energy sources does not allow for a consistent and reliable supply of power throughout the day, thus reducing the overall performance and reliability of the system [13].

Numerous scholars have conducted in-depth research on either photovoltaic maximum power point tracking algorithms or energy optimization management as singular aspects. In terms of the maximum power tracking of PVs, [14] designed a high-gain DC/DC circuit based on the model prediction MPPT algorithm, which is a structure that can provide up to 10 times the voltage gain of the input voltage. in order to improve the convergence speed of the algorithm under uniform irradiation conditions, while [15] proposed an MPPT algorithm based on the golden section search. The authors of [16] proposed a curve-fitting-based hill-climbing MPPT algorithm, which does not depend on the system parameters. These algorithms have good tracking effects in uniform-irradiation environments; however, they fall into the local maximum power point (MPP) in partially shaded environments and cannot effectively track the global MPP. In contrast, [17] proposed an MPPT algorithm that mixes multiple algorithms and the artificial neural network-based MPPT algorithm. In [18], good tracking effects in partially shaded environments are obtained, but these two algorithms are more complex and require high computational and operational capabilities of the MPPT controller. The above studies are limited to the technology of PV maximum power tracking, and a few studies have considered the power quality optimisation of grid-connected PVs, but most of these do not incorporate the actual application requirements, such as base stations.

In the domain of energy optimization management for base stations, research can be classified into two primary categories. One category involves the optimization design and adjustment of base station loads, entailing real-time adjustments and management of power consumption for base stations across various operational environments and demands [19]. Reference [20] introduced a machine learning online algorithm capable of jointly controlling the sleep mode of small base stations and energy sharing. The concept of spatiotemporal energy management was proposed in [5], wherein base stations managed their power consumption based on real-time prices, concurrently adjusting user associations and modifying their power consumption according to price differentials among different base stations. The authors of [21] introduced an active time reduction technique utilizing terminal location information estimated from wireless communication signals to mitigate energy consumption. The other category involved the integration of renewable energy sources such as photovoltaics. The authors of [22] suggested employing battery energy storage systems to assist in addressing the high power consumption issue of base stations through a supplementary renewable energy supply. The authors of [23] utilized the idle capacity of base station energy storage to stabilize the flow of photovoltaic energy towards base stations, thereby reducing the amount of electricity purchased from the grid and consequently lowering energy costs.

Above all, while these studies have achieved certain results in their respective fields, existing research primarily focuses on optimizing the technology itself. The algorithms involved are complex, demanding high computational speed and operational capability. They have not adequately addressed the need for integrating photovoltaic MPPT technology with the optimization of base station power quality, especially in comprehensive control optimization within multi-energy environments.To deal with the issues above, this study addresses the existing system architecture and technical specifications of 5G base station power supply systems. By analyzing the characteristics of photovoltaic cells and the synergy of multi-source microgrid energy, a novel distributed photovoltaic 5G base station DC microgrid structure is proposed. Furthermore, from the perspectives of energy flow and information flow, a distributed photovoltaic 5G base station DC microgrid energy management strategy based on the CF-P&O-INC MPPT algorithm is introduced. This strategy aims to promote the effective utilization of renewable energy, maximize PV energy output, achieve coordinated energy output in various forms in the multi-source power supply system of 5G base stations, and enhance the scalability, distribution efficiency, and security of base stations.

The contributions of this study are summarized as follows:

• Proposing a novel distributed photovoltaic 5G base station power supply topology to mitigate geographical constraints on PV deployment and prevent power degradation in other PV cells due to the decline in output power of a single PV cell. This enhances the overall efficiency of the PV system while reducing the number of power electronic converters in traditional structures.
• Introducing a CF-P&O-INC MPPT algorithm suitable for partially shaded environments. This algorithm can swiftly track the global maximum power point and achieve steady-state operation without oscillations. Under uniform irradiation, it successfully locates the maximum power point with only 12 tracking steps, achieving an efficiency of up to 99.95%. In partially shaded conditions, it takes 20 tracking steps to locate the global maximum power point, with an efficiency of 99.82%.
• Proposing a priority-based energy management strategy that dynamically optimizes and coordinates the energy flow of base stations based on factors such as photovoltaic generation characteristics, energy storage batteries, and base station loads. This strategy ensures the local consumption of photovoltaic energy, thereby alleviating the burden on the grid. Under this strategy, the fluctuation of the direct current bus voltage at base stations during severe load changes is less than 1.875%, with stability restored within 0.07 s.

2. System Model

As shown in Figure 1, in the proposed distributed photovoltaic 5G base station DC microgrid energy management structure, the photovoltaic cells are connected to the base station’s DC bus through MPPT controllers. Energy from the photovoltaic cells is prioritized for consumption, with excess energy stored in the energy storage batteries. In instances where the photovoltaic generation is insufficient to support base station operations, the system automatically supplements energy from both the grid and the energy storage batteries via bidirectional DC/DC converters and grid-connected inverters to maintain the DC bus voltage at 48 V, ensuring the uninterrupted operation of the base station.

In the scheme, each photovoltaic cell is individually connected to a low-power MPPT controller, ensuring that each photovoltaic module operates at its maximum power output by employing MPPT algorithms. This resolves the issue of low efficiency in traditional series-parallel photovoltaic array systems. Additionally, through the Boost circuit in the low-power MPPT controller, which reduces current and increases voltage, the output voltage is adjusted to around 48V, facilitating subsequent energy management steps.

Given that obstructions such as buildings, trees, and cloud cover can cause partial shading on photovoltaic cells, resulting in multiple peaks in their output characteristics, traditional MPPT algorithms may fail to track the global maximum power point due to their tendency to converge to local maxima. Therefore, this scheme proposes a CF-P&O-INC MPPT algorithm to achieve the rapid and steady-state oscillation-free tracking of the maximum power point, while also ensuring the tracking of the global maximum power point under partial shading conditions.

2.1. MPPT Algorithm

2.1.1. CF Method

The polynomial curve fitting method is used to find the maximum power point of the PV cell by fitting the power–voltage (P-V) characteristic curve of the PV cell.

As depicted in Figure 2, the polynomial function has similar nonlinear characteristics to the P-V curve of PVs. Under the conditions of uniform irradiation, the P-V characteristic curve of PV cell has only one extreme value, and the curve monotonically increases to the extreme value first, and then monotonically decreases, which corresponds to the second-order polynomial function. Its higher-order polynomial function possesses multiple extreme points, approximating the P-V characteristics of PV cells under partially shaded conditions, and it is only necessary to fit the curve to the PV voltage range (zero volts-open-circuit voltage), which is simple to operate.

Polynomial function curve fitting equation is as follows:

$$f\left(v\right)=a_0+a_{1}v+a_2{v}^2+\cdots +a_n{v}^n$$

(1)

where $f\left(v\right)$ represents the output power of the photovoltaic cell, v represents the output voltage of the PV cell, n denotes the order of the polynomial function, and $a_n$ denotes the unknown coefficients for each order in the equation.

Employing the least squares method aims to minimize the sum of squares of residuals between the data and the model, thereby optimizing the fitting function to minimize the error between the fitted function and the actual PV P-V characteristic curve. The actual P-V curve can be expressed by the following equation:

$$P\left(v\right)=a_0+a_{1}v+a_2{v}^2+\cdots +a_n{v}^n+e$$

(2)

where e represents the residuals of the model, which are calculated as follows:

$$e=f\left(v\right)-P\left(v\right)$$

(3)

The sum of squares of the residuals is given by Equation (4) below:

$$L=\sum _{i=1}^m{\left(f\left(v_i\right)-P\left(v_i\right)\right)}^2$$

(4)

where m is the number of samples. In order to obtain the best fit coefficients, the sum of squares of the residuals should be minimised. Therefore, for the optimal function, the partial derivation of the sum of squares of the residuals L, as described above, should be equal to 0:

$$\frac{\partial L}{\partial a_j}=2\ast \sum _{i=1}^m\left[\left(a_0+a_{1}v+a_2{v}^2+\cdots +a_n{v}^n-P\left(v_i\right)\right)v_i^j\right]=0$$

(5)

Then, a further simplification of Equation (6) is obtained:

$$a_0\sum _{i=0}^m{v}_i^j+a_1\sum _{i=0}^m{v}_i^{j+1}+a_2\sum _{i=0}^m{v}_i^{j+2}+\cdots +a_n\sum _{i=0}^m{v}_i^{j+n}=\sum _{i=0}^m\left(v_i^j\ast P\left(v_i^j\right)\right)$$

(6)

Substituting $j=0,1,2,\dots ,n$ into the above equation, respectively, we can obtain Equation (7):

$$\left\{\begin{array}{c}m{a}_0+a_1\sum _{i=1}^m{v}_i+a_2\sum _{i=1}^m{v}_i^2+\cdots ++a_n\sum _{i=1}^m{v}_i^n=\sum _{i=1}^m\left(P\left(v_i\right)\right)\\ a_0\sum _{i=1}^m{v}_i+a_1\sum _{i=1}^m{v}_i^2+a_2\sum _{i=1}^m{v}_i^3+\cdots ++a_n\sum _{i=1}^m{v}_i^{n+1}=\sum _{i=1}^m(v_i\ast P\left(v_i\right))\\ a_0\sum _{i=1}^m{v}_i^2+a_1\sum _{i=1}^m{v}_i^3+a_2\sum _{i=1}^m{v}_i^4+\cdots ++a_n\sum _{i=1}^m{v}_i^{n+2}=\sum _{i=1}^m(v_i^2\ast P\left(v_i\right))\\ \cdots \\ a_0\sum _{i=1}^m{v}_i^n+a_1\sum _{i=1}^m{v}_i^{n+1}+a_2\sum _{i=1}^m{v}_i^{n+2}+\cdots ++a_n\sum _{i=1}^m{v}_i^{2n}=\sum _{i=1}^m(v_i^n\ast P\left(v_i\right))\end{array}\right.$$

(7)

Converting Equation (7) its into matrix form gives

$$\mathbf{X}=\left[\begin{array}{ccccc}m& \sum _{i=1}^m{v}_i& \sum _{i=1}^m{v}_i^2& \cdots & \sum _{i=1}^m{v}_i^n\\ \sum _{i=1}^m{v}_i& \sum _{i=1}^m{v}_i^2& \sum _{i=1}^m{v}_i^3& \cdots & \sum _{i=1}^m{v}_i^{n+1}\\ \sum _{i=1}^m{v}_i^2& \sum _{i=1}^m{v}_i^3& \sum _{i=1}^m{v}_i^4& \cdots & \sum _{i=1}^m{v}_i^{n+2}\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ \sum _{i=1}^m{v}_i^n& \sum _{i=1}^m{v}_i^{n+1}& \sum _{i=1}^m{v}_i^{n+2}& \cdots & \sum _{i=1}^m{v}_i^{2n}\end{array}\right]$$

(8)

$$\mathbf{\Theta }=\left[\begin{array}{c}{a}_0\\ a_1\\ a_2\\ \cdots \\ a_n\end{array}\right],\mathbf{Y}=\left[\begin{array}{c}\sum _{i=1}^m\left(P\left(v_i\right)\right)\\ \sum _{i=1}^m(v_i\ast P\left(v_i\right))\\ \sum _{i=1}^m(v_i^2\ast P\left(v_i\right))\\ \cdots \\ \sum _{i=1}^m(v_i^n\ast P\left(v_i\right))\end{array}\right]$$

(9)

Then there is

$$\mathbf{X}\mathbf{\Theta }=\mathbf{Y}$$

(10)

$$\mathbf{\Theta }={\mathbf{X}}^{-\mathbf{1}}\mathbf{Y}$$

(11)

After substituting the samples into the constructed matrices $\mathbf{X}$ and $\mathbf{Y}$, the vector of coefficients of the optimal function $\mathbf{\Theta }$ can be solved from the above equation.

$$\frac{df\left(v\right)}{dv}=0$$

(12)

$$a_{1}v+2a_2{v}^1+3a_2{v}^2+\cdots +n{a}_n{v}^{n-1}=0$$

(13)

The location of the maximum power point can be predicted by taking a first-order derivative of the fitted curve and making it zero in order to find the horizontal coordinates at the point of maximum value of the fitted curve.

2.1.2. Assessing the Presence of Partial Shadowing Conditions

Under the conditions of uniform irradiation, the P-V characteristic curve of photovoltaic cells has only one extreme value, and the second-degree polynomial function has a good fitting effect.

$$f\left(v\right)=a_0+a_{1}v+a_2{v}^2$$

(14)

The proposed algorithm performs curve fitting with a single peak at initialisation to solve the voltage at the maximum power point $v_{cf\_max}$, and uses Equation (4) to calculate the extreme value of the fitted curve $f\left(v_{cf\_max}\right)$, which is obtained according to the Boost circuit input–output relationship:

$$\left\{\begin{array}{c}{v}_{pv}=\frac{1}{1-D}{v}_{out}\\ v_{out}=v_{bus}\\ D=1-\frac{v_{bus}}{v_{pv}}\end{array}\right.$$

(15)

where $v_{pv}$ is the input voltage of the Boost circuit, $v_{out}$ is the output voltage of the Boost circuit, $v_{bus}$ is the DC bus voltage, and D is the duty cycle of the Boost circuit.

From the above equation, the duty cycle corresponding to the extreme value of the fitted curve in the Boost circuit can be calculated.

$$|f\left(v_{cf\_max}\right)-P\left(v_{cf\_max}\right)|>P_{th}$$

(16)

The power threshold $P_{th}$ should be greater than the residual difference between the power value of the fitted curve and the actual PV cell P-V curveat the voltage position of the fitted maximum power point, and it is known from many experiments that the actual threshold should be taken as 5%*$P\left(v_{cf\_max}\right)$ - 10%*$P\left(v_{cf\_max}\right)$.

If the difference between the maximum value $f\left(v_{cf\_max}\right)$ of the fitted curve and the actual power $P\left(v_{cf\_max}\right)$ at the voltage point of the fitted maximum value is less than $P_{th}$, then it can be judged as a single-peak curve under uniform irradiation conditions; otherwise, it is identified as a multi-peak curve.

Under the conditions of partial shading, the probability of double peaks appearing is the highest. In this case, the P-V curve of PVs possesses two maxima and one minima, and from the polynomial property, the fourth-degree polynomial possesses the same property, which can fit the condition of double peaks better. After fitting, the curve will be compared with the threshold value again, and if it is greater than the threshold value, the curve is fitted again; otherwise, the process moves to the second stage.

$$f\left(v\right)=a_0+a_{1}v+a_2{v}^2+a_3{v}^3+a_4{v}^4$$

(17)

From the analysis above, the P-V characteristic curve of a PV cell with n peaks has at least $2n-1$ extreme values, and the fitting curve with the lowest polynomial number of values is

$$f\left(v\right)=a_0+a_{1}v+a_2{v}^2+\cdots +a_{2n}{v}^{2n}$$

(18)

Thus, the range of maximum power points can be determined from the extreme values of the polynomial curve fit:

$$v_{\mathrm{Re}al\_max}\in (v_{cf}-v,v_{cf}+v)$$

(19)

2.1.3. CF-P&O-INC MPPT Algorithm

Aiming at the problems of slow convergence speed, large steady-state oscillations, and ineffective tracking of the global maximum power point in partially shaded environments in MPPT algorithms in existing studies, this paper proposes a CF-P&O-INC MPPT algorithm to solve the above problems.

The first stage of the proposed algorithm uses polynomial curve fitting to quickly determine the range of the maximum power point, and the second stage uses an improved P&O algorithm to quickly locate the maximum power point position.

The P&O algorithm exhibits good stability, based on periodically perturbing the output voltage of photovoltaic cells at certain intervals, inducing increments or decrements. Simultaneously, the output power is observed to ascertain the direction of change, serving as the basis for determining subsequent variations in control signals. Traditional P&O algorithms employ fixed step sizes, where larger steps lead to faster convergence but greater steady-state oscillations, and smaller steps result in reduced steady-state oscillations, albeit with slower convergence. An adaptive P&O algorithm is proposed to achieve rapid convergence in the second stage.

If $\mathrm{\Delta }P>0$, it indicates that the power at the current moment is greater than that at the previous moment. The direction of voltage adjustment is correct, and the perturbation can continue in the original direction. If the perturbation direction remains unchanged on three consecutive occasions, the step size is increased.

$$\mathrm{\Delta }D\left(t\right)=\mathrm{\Delta }D(t-1)$$

(20)

If $\mathrm{\Delta }P<0$, the voltage is adjusted in the wrong direction, the direction of the perturbation needs to be changed and the step size reduced immediately.

$$\frac{\mathrm{\Delta }I}{\mathrm{\Delta }V}=-\frac{I}{V}$$

(21)

In the third stage, the fine step size further tracks the MPP and determines whether the photovoltaic cells are operating at the maximum power point. When the condition stipulated in Equation (21) is met, and the variation in output conductance equals the negative value of the output conductance, the photovoltaic cell panel operates at the maximum power point. At this juncture, perturbations cease, thereby achieving a steady state without oscillation.

The change in solar irradiance causes a shift in the position of the maximum power point, and the irradiance is proportional to the current [24].

$$I_{pv}=\alpha S$$

(22)

Based on the above formula, the magnitude of changes can be determined, where S represents irradiance. When irradiance changes slightly, the output current $I_{pv}$ of the photovoltaic cells varies slowly, resulting in a slight shift in the MPP position. The proposed algorithm operates in the third stage with smaller steps for fine-tuning, dynamically tracking the changing MPP. When irradiance changes moderately, the algorithm switches back to the second stage, employing a P&O algorithm with adaptive step size for disturbance handling. For significant changes in irradiance, causing a large shift in the MPP position, the algorithm reinitializes curve fitting to quickly respond to changes in external environmental conditions. As shown in Figure 3.

2.2. DC Bus Voltage Control

Voltage control is the core of energy management in DC microgrids for 5G base stations, where maintaining voltage stability is paramount. In the multi-source system of photovoltaic 5G base station DC microgrids, the fluctuation in PV output power due to factors such as solar irradiance and temperature results in voltage instability. The output voltage of the PV side does not remain stable at 48V but drifts around this value. Additionally, fluctuations in base station load power and the presence of various types of DC/DC and AC/DC converters within the base station necessitate the design of effective and rational control techniques.

To maintain the stability of the DC bus voltage, it is imperative to uphold power balance within the system, ensuring that the power injected into the DC bus equals the power consumed by the DC loads. The constraint for DC bus power balance is articulated as follows:

$$P_{supply}\left(k\right)=\sum _{i=1}^{\mathrm{n}}{P}_{pv\left(i\right)}\left(k\right)+P_{grid}\left(k\right)+P_{bat}\left(k\right)$$

(23)

$$P_{load}\left(k\right)={\eta }_1{P}_{AAU}\left(k\right)+{\eta }_2{P}_{BBU}\left(k\right)+{\eta }_3{P}_{oth}\left(k\right)$$

(24)

where $P_{supply}\left(k\right)$ is the sum of power provided by all power sources and $P_{load}\left(k\right)$ is the total power demanded by the load. ${\displaystyle \sum _{i=1}^{\mathrm{n}}}{P}_{pv\left(i\right)}\left(k\right)$, $P_{grid}\left(k\right)$, $P_{bat}\left(k\right)$, $P_{AAU}\left(k\right)$, $P_{BBU}\left(k\right)$, and $P_{oth}\left(k\right)$ represent the power of the PVs, grid, storage batteries, AAUs, BBUs, and other DC loads at the moment k, and ${\eta }_1$, ${\eta }_2$, and ${\eta }_3$ are the loss coefficients in the transmission of power from AAUs, BBUs, and other DC loads, respectively.

$$P_{dc\_net}\left(k\right)=P_{supply}\left(k\right)-P_{load}\left(k\right)$$

(25)

$$P_{dc\_net}\left(k\right)=V_{dc}\left(k\right)\ast i_{e}k)$$

(26)

The voltage control loop maintains the given reference DC bus voltage by adjusting the net current $i_e$ given by Equation (26).

$$V_e\left(k\right)=V_{dc\_ref}\left(k\right)-V_{dc}\left(k\right)$$

(27)

$$i_e=K_{pe}{V}_e\left(k\right)+K_{ie}\int V_e\left(k\right)dk$$

(28)

where $V_e\left(k\right)$ is the difference between the DC bus reference voltage and the actual voltage at moment k, while $K_{pe}$ and $K_{ie}$ are the proportionality and integration constants of the voltage control loop, respectively.

Considering the PV current $I_{pv}$, the net current can be expressed as Equation (29) and the average current component is extracted by a low-pass filter with a cut-off frequency of $(1/T_e)$, given by Equation (30).

$$i_{e2}\left(k\right)=i_e\left(k\right)-I_{PV}\left(k\right)$$

(29)

$$i_{e1}\left(k\right)=\frac{1}{1+s{T}_e}{i}_{e2}\left(k\right)$$

(30)

Subsequently, the energy management strategy determines the magnitude of energy supplied to the battery and the grid based on the average current component $i_{e1}\left(k\right)$, the power supply conditions of the base station, and the state of charge of the battery. Collectively, these factors ensure the maintenance of the bus voltage within the standard voltage range for the base station.

2.3. Priority-Based Energy Management Strategy

The priority-based energy management strategy determines the direction of energy flow based on the operation index of the internal power system of the base station, the power generation of the PV array, the ability of the power grid to inject and absorb power, and the charging state of the energy storage battery. Three operating modes are defined based on the PV power generation and the power consumption of the base station: the power shortage mode ($P_{pv}<P_{dc\_load}$), the balance mode ($P_{pv}=P_{demand}$), and the residual power mode ($P_{pv}\ge P_{demand}$)

(1)

Power shortage mode

In the power shortage mode, all electrical energy generated by the photovoltaic cells is directed towards the DC load, while energy storage batteries serve as emergency backup power sources for the base station. In the event of external power failure or interruption, these batteries provide critical electrical support, ensuring uninterrupted operation of the base station. Due to the flow-sensitive nature of power loads in 5G base stations, internal backup batteries gradually accumulate reserve capacity over time. This reserve capacity is adjustable and can be utilized as a flexible resource within the power system. Consequently, the deficient portion of power is primarily supplied by the energy storage batteries. If the adjustable capacity of the energy storage batteries is depleted or if the combined power output is insufficient to meet the demands of the base station, the remaining power shortfall is supplemented by the grid.

The capacity of backup batteries can dynamically be divided into two parts: the first part can be used for emergency purposes to guarantee base station reliability and is not involved in energy management. The other part of the capacity is available for flexible scheduling. Both portions of capacity vary with the base station load power, necessitating an assessment to determine the minimum reserved capacity $R_b\left(t\right)$ required to maintain reliability requirements, thereby enabling the determination of adjustable capacity. $R_b\left(t\right)$ is calculated based on power load $L_b\left(t\right)$ and emergency duration $D_b$ [25].

$$R_b\left(t\right)=\underset{t}{\overset{t+D_b}{\int }}{L}_b\left(\tau \right)d\tau$$

(31)

(2)

Balance mode

The system is in the balanced mode when the power output from the PV cells is equal to the power consumed by the loads inside the base station. In this mode, all the power required by the DC loads is provided by the PV cells, while the grid connection and energy storage batteries are responsible only for maintaining the stability of the DC bus voltage.

(3)

Residual power mode

In this mode, the electrical energy required by both the DC and AC loads is supplied by the photovoltaic system. Any surplus power is first directed to the energy storage batteries, and then, if available, is fed into the grid. This ensures that the electricity generated by the photovoltaic system is primarily utilized on site, thereby alleviating the burden on the grid.

3. Simulink Simulation and Analysis

In order to verify the effectiveness of the proposed MPPT algorithm and energy management strategy, a distributed PV 5G base station DC microgrid model was constructed in MATLAB/Simulink (version R2022b), as shown in Figure 4. The topology of this model consists of PV cells, MPPT controller, DC bus, energy storage battery, bidirectional DC/CD converter, grid, bidirectional AC/DC converter, DC load and AC load.

3.1. Performance of the Proposed MPPT Algorithm under Standard Test Conditions

We tested the proposed CF-P&O-INC MPPT algorithm under conditions of uniform irradiation to validate its performance regarding convergence speed, steady-state oscillation, and other relevant aspects. The experimental parameters are depicted in

Table 1. Photovoltaic cell parameters.

Electrical Characteristics	Value
Maximum power	602.26 W
Open circuit voltage	44.52 V
Short circuit current	18.3 A
PV output voltage at MPP	36 V
PV output current at MPP	16.74 A

.

During the simulation, the temperature was set to be constant at room temperature (25 °C), the irradiance was 1 $\mathrm{kW}/{\mathrm{m}}^2$, and the tracking rate of the algorithm was 3 ms. The effect of the CF-P&O-INC MPPT algorithm under uniform irradiation conditions is given in Figure 5 and Figure 6, including the output power of the PV cell, the duty cycle of the MPPT controller, and the current vs. voltage.

From Figure 5 and Figure 6, it can be seen that the tracking process of the proposed algorithm can be divided into three stages. In the first stage, the voltage and current data of the three points that are distributed far away are sampled, and then polynomial curve fitting is performed to calculate the location of the theoretical maximum power point, which is successfully localised to the nearby area of the maximum power point in the fourth tracking. In the second stage, the maximum power point is tracked using the variable step size P&O algorithm, which uses a total of eight tracking steps. The INC stage verifies the localised MPP to bring the system to a steady state. Thus, the proposed algorithm successfully locates the maximum power point location using a total of 12 tracking steps with a single tracking interval of 3 ms, resulting in a final total duration of 36 ms. The output power in steady state is 601.6 W, while the maximum output power of the PV cell is 602.64 W. The MPPT efficiency is as high as 99.95%.

3.2. Performance of the Proposed MPPT under Step Change in Solar Irradiance

During the simulation, to test the performance of algorithms under different irradiance conditions, the temperature was set constant at 25 °C. Irradiance initially was 1 $\mathrm{W}/{\mathrm{m}}^2$, and at 1 s, it abruptly changed to 600 $\mathrm{W}/{\mathrm{m}}^2$, and at 2 s, it changed to 800 $\mathrm{W}/{\mathrm{m}}^2$. Simultaneously, to test the performance of algorithms under different temperature conditions, irradiance was set constant at 1000 $\mathrm{W}/{\mathrm{m}}^2$, and the temperature was initially set at 25 °C. At 1 s, temperature rapidly increased to 45 °C, and at 2 s, it rapidly decreased to 35 °C. Under the same simulation conditions, the proposed algorithm was tested and compared with traditional P&O and INC algorithms. The tracking results of these three algorithms are shown in Figure 7.

From Figure 7a, it can be seen that the proposed algorithm at the initial time tracks at the maximum power point at 36 ms, while the P&O and INC take 192 ms and 330 ms to reach the steady state, respectively. For the first step change in irradiance, the proposed algorithm, P&O and INC algorithms took 27 ms, 30 ms and 96 ms, respectively, and for the second step change in irradiance, the three algorithms took 21 ms, 27 ms and 66 ms, respectively. From Figure 7b, it can be seen that during the first temperature change, the proposed algorithm, the P&O algorithm, and the INC algorithm took 36 ms, 48 ms, and 123 ms, respectively, to reach the steady-state. During the second temperature change, the three algorithms took 12 ms, 15 ms, and 42 ms, respectively. Overall, the proposed algorithm demonstrates faster response times and more stable performance under varying irradiance and temperature conditions, making it suitable for practical applications requiring rapid adaptation to dynamic environmental changes.

3.3. Performance of the Proposed MPPT in a Partially Shaded Environment

To validate the performance of the proposed algorithm under partial shading conditions, we simulated scenarios using three connected PV modules subjected to varying irradiance levels. Three groups of simulations were conducted. In the first group, the PV modules were exposed to irradiance levels of (1000, 1000, 600) $\mathrm{W}/{\mathrm{m}}^2$ respectively, at a temperature of 25 °C. Under these conditions, the global maximum power output of the PV modules was 392.8 W, with a local maximum of 320.4 W. In the second group of tests, the irradiance levels remained (1000, 1000, 600) $\mathrm{W}/{\mathrm{m}}^2$, but the temperature was increased to 45 °C, resulting in a global maximum power output of 361 W and a local maximum of 298.7 W. In the third group, the modules were exposed to irradiance levels of (850, 850, 580) $\mathrm{W}/{\mathrm{m}}^2$ at 25 °C, yielding a global maximum power output of 335.9 W and a local maximum of 306.2 W. The effectiveness of the proposed algorithm is illustrated in Figure 8.

In Figure 8, it is evident that across the three test groups, the proposed algorithm converges to the global maximum power point at 60 ms, 72 ms, and 81 ms, respectively, demonstrating steady-state operation without oscillations. The average output powers in steady state are 392.6 W, 360.1 W, and 335 W, with MPPT efficiencies of 99.82%, 99.75%, and 99.73%, respectively.

In contrast, under identical experimental conditions, the P&O and INC algorithms converge to local maximum power points, failing to effectively track the global maximum power point. In the three test sets, the P&O algorithm stabilizes at local maximum power points at 117 ms, 147 ms, and 135 ms, exhibiting significant oscillations in steady state, with average output powers of 319.45 W, 297.8 W, and 305.1 W, respectively. The Incremental Conductance algorithm converges to local maximum power points at 204 ms, 243 ms, and 207 ms, showing minor oscillations in steady state, with average output powers of 320 W, 298.1 W, and 305.7 W, respectively.

From the simulation results depicted in Figure 8a,b, the CF-P&O-INC MPPT algorithm effectively identifies the global maximum power point under partial shading conditions and different temperatures, achieving steady-state operation without oscillations. Conversely, in the comparison presented in Figure 8a,c, the proposed algorithm exhibits robust tracking capability across varying irradiance levels. In contrast, the P&O and INC algorithms are limited by local optima and fail to locate the global maximum power point.

3.4. Energy Management

(1)

Power shortage mode

In this simulation mode, the battery capacity is 100 Ah, and the photovoltaic output power is 2800 W. The initial power demand of the DC load is 4000 W, which decreases to 3200 W at 5 s and increases to 3600 W at 10 s.

When the PV power is insufficient to meet the load demand and the battery SOC is high, the output power of each module is depicted in Figure 9. In this scenario, the battery supplements the insufficient power. Both the photovoltaic system and the battery collectively supply power to the base station load, while the grid does not provide power. At 5 s, when the required load power of the base station decreases to 3200 W, the output power of the battery decreases accordingly. During this period, the DC bus voltage exhibits fluctuations of ±0.7 V, equivalent to 1.46% of the rated voltage, and stabilizes at 47.9 V after 0.1 s. Similarly, at 10 s, when the required load power of the base station increases to 3600 W, the output power of the battery rises accordingly. The DC bus voltage then experiences fluctuations of ±0.4 V, amounting to 0.625% of the rated voltage, and returns to a stable value of 47.9 V after 0.09 s, meeting the requirement specified in IEEE Std 1564™-2014 [26] for the normal operating range of the bus voltage to be within 10% of the rated voltage.

When the PV power is insufficient to meet the load demands and the battery’s SOC is low, the output power of each module is illustrated in Figure 10. In this scenario, with the battery lacking dispatchable capacity, it cannot supply power to the load. Consequently, all energy generated by the photovoltaic system is directed towards the DC load, with the grid supplementing the remaining energy requirements. Both sources collectively power the base station. At the 5th second, when the required load power of the base station decreases to 3200 W, the output power of the battery decreases. During this interval, the DC bus voltage exhibits fluctuations of ±0.9 V, equivalent to 1.875% of the rated voltage, returning to a steady state of 47.8 V after 0.06 s. Subsequently, at the 10th second, when the required load power of the base station increases to 3600 W, the output power of the battery rises. The DC bus voltage then experiences fluctuations of ±0.5 V, also amounting to 1.04% of the rated voltage, returning to a steady state of 47.8V after 0.04 s.

(2)

Balance mode

In this simulation model, the PV output power is 3600 W, which is equal to the power of the base station load.

The output power of each module in the balance mode is shown in Figure 11, where the PV output power flows completely to the base station load, the battery and the grid is responsible for maintaining the stability of the DC bus voltage. At this time, the DC bus voltage being stable at 47.8 V.

(3)

Residual power mode

In this simulation scenario, the photovoltaic output power is 4600 W, while the initial power demand of the DC load is 4000 W, decreasing to 3200 W at the 5th second, and increasing to 3600 W at the 10th second.

In the surplus electricity mode, the output power of each module is illustrated in Figure 12, with the DC load’s power consumption supplied by the photovoltaic system. Excess electrical energy is primarily directed towards the battery, followed by the grid. At the initial moment, the current flowing from the photovoltaic system to the battery is 12.3 A. Subsequently, after the 5th second, as the power demand of the base station decreases to 3600 W, the surplus power from the photovoltaic system increases, resulting in an augmentation of the current flowing into the battery. This current surpasses the maximum charging current threshold of the battery, set at 20 A, causing the excess power to be directed towards the grid. At the 5th and 10th seconds, corresponding to variations in the base station’s power demand, the DC bus voltage exhibits fluctuations of ±0.7 V and ±0.3 V, respectively, returning to a steady state after 0.07 s and 0.06 s, respectively.

4. Conclusions

This study presents a novel solution for DC microgrid systems in 5G base stations, addressing the challenge of high power consumption by effectively increasing PV generation through the proposed structure and MPPT algorithm. Furthermore, it employs an energy management strategy to facilitate the coordinated operation of multiple energy sources, ensuring a stable and reliable power supply for base stations.

Firstly, the distributed PV structure significantly reduces geographical constraints on PV deployment, mitigating power degradation issues in series-parallel PV arrays and thereby enhancing overall system efficiency. Secondly, a CF-P&O-INC MPPT algorithm is introduced, featuring rapid convergence, high precision, and stable operation without oscillation. Under uniform irradiance, it achieves maximum power point tracking within 36 ms with an efficiency of 99.95%. In partially shaded conditions, it converges to the global maximum power point within 60 ms with an efficiency of 99.82%, effectively improving the efficiency of individual PV cells. Finally, building upon this structure and MPPT algorithm, a priority-based distributed PV 5G base station DC microgrid energy management strategy is proposed. This strategy ensures that the DC bus voltage fluctuates by less than 1.875% during rapid changes in station load, returning to stability within 0.07 s. This strategy facilitates various forms of energy coordination output in 5G base station multi-source power supply systems, enhances the on-site utilization of PV energy, reduces adverse effects such as harmonic pollution and load imbalance caused by PV grid feeding, and strengthens base station scalability, distribution efficiency, and safety.