Dynamic energy distribution optimization method and system for photovoltaic energy storage charging station

Abstract

The invention provides a dynamic energy distribution optimization method and system for a photovoltaic energy storage charging station, is applied to the technical field of power supply and distribution control, and aims at the complexity of multi-source heterogeneous data in a photovoltaic energy storage charging scene and based on an industrial big data analysis technology, and comprises the following steps: collecting multi-source data to construct a variable set, and inputting a model to obtain a supply and demand prediction value; generating an electric quantity flow direction matrix by adopting multi-particle strategy global optimization, generating dynamic boundary verification based on a battery health index, repairing an out-of-limit path by utilizing linear programming, and finally generating a control instruction to drive a power device to execute; according to the method, industrial big data value can be excavated to realize global economy and local safety collaborative optimization of energy distribution, system loss is effectively reduced, the service life of an aged battery is prolonged, control failure caused by sudden working conditions is avoided, and the operation stability and response real-time performance of the micro-grid are improved.

Description

Dynamic energy distribution optimization method and system for photovoltaic energy storage charging stations Technical Field

[0001] This invention relates to the field of new energy microgrid control technology, and in particular to a dynamic energy distribution optimization method and system for photovoltaic energy storage charging stations based on industrial big data analysis. Background Technology

[0002] As a component of new energy microgrids, photovoltaic energy storage charging stations undertake multiple functions, including local consumption of photovoltaic power generation, regulation of energy storage, and supply of charging loads for electric vehicles.

[0003] As photovoltaic energy storage charging stations accumulate operating time, the system generates massive amounts of historical operating logs, battery status data, and environmental monitoring data, constituting typical industrial big data. However, existing energy management systems often lack in-depth mining and effective utilization of this industrial big data, failing to extract the evolutionary patterns of battery aging or deep characteristics of user load from massive historical data, making it difficult for control strategies to achieve self-evolution. Existing energy management systems typically employ energy allocation strategies based on rule-based logic or single optimization algorithms. Among these, rule-based methods have fast response speeds but struggle to achieve comprehensive optimization of economic efficiency and equipment lifespan; while global optimization algorithms such as particle swarm optimization can improve overall system efficiency, their computation cycles are long, making it impossible to respond promptly to real-time safety constraints such as changes in battery health status or sudden load changes. Furthermore, traditional solutions generally lack deep perception of the health status of energy storage batteries during scheduling, allocating power solely based on state of charge, ignoring aging characteristics such as battery internal resistance growth and capacity decay. This leads to aging batteries being prone to excessive temperature rise, voltage exceeding limits, and even thermal runaway risks under high-power operation. When the global optimization result violates the security boundary, existing systems often adopt a replanning strategy, which causes control delays and wastes computing resources, resulting in a technical bottleneck where real-time performance and security are difficult to balance.

[0004] As battery service life increases, their electrochemical performance deteriorates more rapidly. How to achieve efficient and long-life operation while ensuring the dynamic safety of the system has become a challenge for energy management of photovoltaic energy storage charging stations.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a dynamic energy allocation optimization method and system for photovoltaic energy storage charging stations, aiming to solve the problems of insufficient battery health status perception, poor real-time performance of global optimization algorithms, and lag in control strategy response under sudden safety constraints in the prior art. By introducing a dynamic safety power boundary based on battery health indicators and combining global optimization and local linear repair mechanisms, the present invention achieves active protection of aging batteries, improves the system's response capability under second-level disturbances, and ensures that the energy allocation scheme continues to operate stably under the premise of meeting physical safety constraints.

[0007] This invention provides a dynamic energy allocation optimization method for photovoltaic energy storage charging stations, comprising: utilizing a sensor network deployed at photovoltaic panels, charging piles, and energy storage batteries to collect real-time power generation, load demand, and battery remaining capacity and health indicator data; mapping the real-time power generation, load demand, battery remaining capacity, and health indicator data into a comprehensive environmental variable set; inputting the comprehensive environmental variable set into a prediction model to obtain the photovoltaic supply forecast and charging demand forecast for the future time period; and generating an initial allocation demand vector based on the photovoltaic supply forecast and charging demand forecast; and using a multi-particle path search strategy to globally optimize the initial allocation demand vector with the goal of minimizing system energy transmission loss and battery life loss. The optimization process generates a power flow matrix that includes the energy distribution ratio between photovoltaic power generation units, energy storage battery units, and charging pile units. It then verifies whether the planned battery discharge rate in the power flow matrix exceeds the dynamic safety power boundary generated in real-time based on health indicator data. If the battery discharge rate exceeds the dynamic safety power boundary, it locks the flow paths in the power flow matrix that have not exceeded the limit. For the paths that exceed the limit, it uses a linear programming algorithm to redistribute the power difference among spare nodes with power adjustment margins within the system, obtaining a corrected power flow matrix. Finally, it converts the corrected power flow matrix into current control commands and sends these commands to the photovoltaic inverter, energy storage bidirectional converter, and charging pile controller, driving the power devices to perform energy distribution.

[0008] In some optional embodiments, the process of mapping real-time power generation, load demand power, battery remaining capacity, and health index data into a comprehensive set of environmental variables specifically includes: synchronously collecting the irradiance and backsheet temperature of the photovoltaic panel, the access voltage and requested current of the charging pile, and the single-cell terminal voltage and real-time temperature of the energy storage battery; using a feature extraction algorithm to extract the supply intensity factor characterizing the output capacity of the photovoltaic array and the health degradation factor characterizing the degree of electrochemical aging of the battery from the collected data; and combining the supply intensity factor, health degradation factor, and real-time collected power data to form a comprehensive set of environmental variables.

[0009] In some optional embodiments, the process of generating the initial allocation demand vector specifically includes: calculating the difference between the photovoltaic supply forecast and the charging demand forecast; when the difference shows that the supply is less than the demand, determining whether the remaining battery capacity is higher than the minimum maintenance capacity; if the remaining battery capacity is higher than the minimum maintenance capacity, determining the discharge power gap that needs to be provided by the energy storage battery based on the difference, and writing the discharge power gap as a constraint condition into the initial allocation demand vector.

[0010] In some optional embodiments, the process of global optimization using a multi-particle path search strategy specifically includes: initializing a particle swarm, where the position vector of each particle represents a possible power flow matrix; constructing a fitness function, which includes an energy transfer efficiency term, a grid interaction cost term, and a battery temperature penalty term; during the search process, calculating the expected battery temperature rise value corresponding to each particle in real time, marking the positions of particles whose expected battery temperature rise value exceeds a preset safety value as invalid solutions, and guiding the particle swarm to converge towards the position with the optimal fitness function value and the expected battery temperature rise value within the safety range.

[0011] In some optional embodiments, the power flow matrix is ​​a two-dimensional matrix, where the rows of the two-dimensional matrix represent energy output sources, the columns of the two-dimensional matrix represent energy receiving ends, and the element values ​​in the two-dimensional matrix represent the power transmitted; the global optimization process also includes satisfying the energy conservation constraint, which is that the sum of the elements in each column of the two-dimensional matrix is ​​equal to the current power demand of the corresponding energy receiving end.

[0012] In some optional embodiments, the process of generating the dynamic safe power boundary includes: establishing a nonlinear mapping relationship between health index data and the maximum allowable charge / discharge rate; when the health index data indicates that the battery internal resistance increases or the capacity decays, proportionally shrinking the value of the maximum allowable charge / discharge rate; and weighting and correcting the shrunken rate in combination with the real-time temperature of the energy storage battery to obtain the dynamic safe power boundary at the current moment.

[0013] In some optional embodiments, the process of redistributing power differences among standby nodes with power regulation margins within the system using a linear programming algorithm specifically includes: calculating the difference between the planned power and the dynamic safe power boundary in the over-limit path to obtain the power to be transferred; identifying standby battery cells or grid connection nodes whose current load rate is not saturated in the power flow matrix; and using the simplex method to calculate the optimal proportion of the power to be transferred to the standby battery cells or grid connection nodes with the objective function of minimizing the total allocation deviation.

[0014] In some optional embodiments, after sending the current control command, the method further includes a real-time smoothing step: acquiring actual load fluctuation data of the charging pile at millisecond intervals; calculating the instantaneous deviation between the actual load fluctuation data and the corrected power flow matrix; and using a PID controller to generate a compensation signal for the bidirectional converter of the energy storage based on the instantaneous deviation.

[0015] In some alternative embodiments, the process of generating a compensation signal for the bidirectional energy storage converter using a PID controller includes performing adaptive gain adjustment based on battery state: acquiring the dynamic impedance parameters of the energy storage battery in real time; reducing the proportional gain and derivative gain coefficients of the PID controller when the dynamic impedance parameters increase to slow down the rate of change of the output current of the bidirectional energy storage converter; and restoring the gain coefficient of the PID controller to improve the response speed to load fluctuations when the dynamic impedance parameters are within the normal range.

[0016] In some optional embodiments, the real-time smoothing step further includes dead-zone control logic: setting a response dead-zone threshold for power fluctuations; outputting a compensation signal through the PID controller only when the absolute value of the instantaneous deviation is greater than the response dead-zone threshold; and keeping the output power of the energy storage bidirectional converter constant when the instantaneous deviation is within the response dead-zone threshold, allowing the grid side to naturally absorb small fluctuations.

[0017] In some optional embodiments, the method further includes long-term evolution and parameter update steps: storing the comprehensive set of environmental variables generated daily, the corrected power flow matrix, and the deviation data after actual execution as historical scene logs; calculating the cosine similarity between the environmental features of the current date and the environmental features of historical dates recorded in the historical scene logs; selecting the historical date with the highest similarity as a reference scene, and extracting the correction parameters under the reference scene.

[0018] In some optional embodiments, the process of extracting and applying correction parameters under the reference scenario includes: analyzing the temperature rise curve and voltage fluctuation curve of the battery in the reference scenario; if the battery in the reference scenario experiences overheating or overvoltage during a specific period, then reducing the setting value of the dynamic safe power boundary for the corresponding period of the current date; if photovoltaic curtailment occurs in the reference scenario, then adjusting the fitness function weight in the global optimization process and increasing the reward weight for photovoltaic consumption.

[0019] In some optional embodiments, the health indicator data includes at least the battery's state of health (SOH) and internal resistance growth rate; the step of verifying the battery discharge rate planned in the power flow matrix further includes: when the internal resistance growth rate exceeds a preset aging threshold, forcibly limiting the charging power flowing to the battery cell in the power flow matrix to prevent the terminal voltage from being artificially high due to excessive internal resistance.

[0020] In some alternative embodiments, the method is applied to a charging station system comprising multiple parallel energy storage battery clusters; the power flow matrix plans the charging and discharging power separately for each independent energy storage battery cluster; the modified power flow matrix enables battery clusters with poor health indicators to undertake smaller power tasks, and battery clusters with better health indicators to undertake larger power tasks.

[0021] In some optional embodiments, the data acquisition frequency of the sensor network is higher than the generation frequency of the power flow matrix; the prediction model adopts a sliding window mechanism, continuously updates the prediction values ​​for future time periods using the latest acquired data, and triggers the re-execution of the multi-particle path search strategy when the prediction deviation is detected to exceed the replanning threshold.

[0022] This invention provides a dynamic energy allocation optimization system for photovoltaic energy storage charging stations, comprising: a multi-dimensional sensing and acquisition module, deployed at the station site, configured to synchronously acquire photovoltaic power generation, charging pile load demand, and BMS status data of the energy storage battery through a sensor network, and map the photovoltaic power generation, charging pile load demand, and BMS status data into a comprehensive set of environmental variables; a supply and demand forecasting and planning module, connected to the multi-dimensional sensing and acquisition module, configured to run a preset forecasting model to generate trend forecasts of photovoltaic supply and charging demand, and produce an initial allocation demand vector accordingly; and a global path optimization module, connected to the supply and demand forecasting and planning module, configured to execute a multi-particle path search algorithm, and optimize the energy allocation of a photovoltaic power generation, charging pile load demand, and BMS status data into a comprehensive set of environmental variables; a supply and demand forecasting and planning module, connected to the supply and demand forecasting and planning module, configured to execute a multi-particle path search algorithm, and optimize the energy allocation of a photovoltaic power generation, charging pile load demand, and BMS status data into a comprehensive set of environmental variables; and a global path optimization ... Under the premise of sufficient energy conservation, with the goal of minimizing the total system loss, the power flow matrix among the photovoltaic, battery, and charging pile is output. The safety boundary correction module, connected to the global path optimization module, is configured to verify in real time whether the power flow matrix exceeds the dynamic safety power boundary constructed based on battery health indicators. When the boundary is detected to be exceeded, this module is configured to use a linear programming algorithm to recalculate the power allocation ratio among the standby nodes that have not exceeded the limit, and generate a corrected power flow matrix. The hardware execution control module, connected to the safety boundary correction module, is configured to parse the corrected power flow matrix into underlying drive signals and distribute them to the photovoltaic inverter, energy storage bidirectional converter, and charging pile control unit.

[0023] In some optional embodiments, the safety boundary correction module includes a health-power mapping unit; the health-power mapping unit stores a nonlinear lookup table of battery state of health (SOH) and maximum allowable charge-discharge rate (C-rate); the unit is configured to automatically narrow the numerical range of the dynamic safety power boundary by querying the nonlinear lookup table when the battery state of health (SOH) value decreases, and input the narrowed boundary value as a constraint condition into the solver of the linear programming algorithm.

[0024] In some optional embodiments, the hardware execution control module embeds a variable gain damping controller; the variable gain damping controller is configured to monitor the instantaneous fluctuations of the actual load during execution and adjust the differential gain parameter of the PID closed-loop control loop according to the real-time dynamic internal resistance value of the energy storage battery; when the real-time dynamic internal resistance value exceeds the aging baseline, the variable gain damping controller automatically lowers the differential gain parameter to suppress sudden changes in current commands for high internal resistance batteries.

[0025] In some optional embodiments, the system further includes a self-evolutionary learning engine; the self-evolutionary learning engine is connected to the multi-dimensional perception acquisition module and the global path optimization module, and is configured to maintain a scene database containing historical environmental features and optimal control strategies; the engine is configured to periodically calculate the vector similarity between the current environmental features and the historical records in the scene database, and when a match is successful, inject the corresponding historical optimal control parameters into the global path optimization module as initial population parameters.

[0026] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention.

[0027] The present invention provides a dynamic energy distribution optimization method and system for photovoltaic energy storage charging stations, which has the following beneficial effects: By integrating battery health status and dynamic safety boundaries, the present invention achieves active protection for aging batteries, avoiding the risks of voltage exceeding limits and thermal runaway caused by increased internal resistance. It adopts a dual-layer control architecture combining global optimization and local repair, ensuring system economy while meeting real-time safety requirements. The introduction of a variable gain damping control strategy improves the operational stability of high-internal-resistance batteries under power surges. Combined with a historical scenario self-evolution mechanism, the system possesses continuous optimization capabilities, significantly extending the lifespan of energy storage equipment and improving the local energy consumption rate. Attached Figure Description

[0028] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0029] Figure 1 is a flowchart of a dynamic energy allocation optimization method for a photovoltaic energy storage charging station according to an embodiment of the present invention; Figure 2 is a structural schematic diagram of a dynamic energy allocation optimization system for a photovoltaic energy storage charging station according to an embodiment of the present invention. Detailed Implementation

[0030] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0031] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0032] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined. Therefore, the actual execution order may change depending on the specific circumstances.

[0033] In multi-source heterogeneous energy systems, dynamic energy allocation must balance global economic efficiency with local security. By mapping real-time state data of the physical system into calculable comprehensive variables, a quantitative characterization of complex operating conditions can be achieved. Supply and demand forecasting based on this variable set allows for prediction of system operating trends over time, providing a basis for proactive control. The global optimization process, aiming to minimize transmission and equipment losses, essentially searches for the optimal power distribution path that satisfies energy conservation constraints in a high-dimensional solution space. Its feasibility depends on accurate modeling of the system's internal loss mechanisms. However, global optimization results may violate device-level safety boundaries, especially when battery health deteriorates, leading to a decrease in electrochemical response. Fixed-parameter control strategies are prone to voltage exceedances or excessive temperature rises. To address this, a dynamic constraint mechanism based on health indicators is introduced, allowing safety boundaries to adjust in real-time with device status, thus transforming electrochemical aging effects into mathematical constraints that can participate in optimization. Furthermore, a local repair strategy replaces global recalculation, correcting infeasible solutions without interrupting control continuity and ensuring that scheduling commands remain within the feasible domain. The final execution stage faces power fluctuations on the order of seconds, at which point the controller's dynamic response directly impacts system stability. By establishing a negative correlation between control gain and battery dynamic impedance, the control strength adaptively decreases as internal resistance increases, thereby suppressing the rate of current change and preventing uncontrollable instantaneous voltage drops in high-resistance batteries. This mechanism achieves layer-by-layer decoupling and closed-loop feedback from macroscopic optimization to microscopic response, ensuring both long-term operational economy and meeting the safety requirements of transient processes.

[0034] As shown in Figure 1, this embodiment of the invention provides a dynamic energy allocation optimization method for photovoltaic energy storage charging stations. The method includes the following steps: Step S100: Collect multi-source operating data and generate a comprehensive environmental variable set.

[0035] In one embodiment, a sensor network deployed in photovoltaic arrays, charging piles, and energy storage battery systems collects real-time data on power generation, load demand, remaining battery capacity, and indicators reflecting battery health. The collected data, after synchronous processing, is integrated into a high-dimensional set of comprehensive environmental variables to characterize the system's current operating condition. In other optional embodiments, the comprehensive environmental variable set can be constructed using data fusion strategies of different dimensions, such as weighted superposition, feature embedding, or vector concatenation.

[0036] Step S200: Based on the comprehensive set of environmental variables, perform supply and demand forecasting and generate an initial allocation demand vector.

[0037] The aforementioned set of comprehensive environmental variables is input into a pre-set prediction model to obtain predicted values ​​for photovoltaic power generation capacity and charging load demand over a future period. Based on the matching relationship between these two values, the overall energy surplus / deficit status of the system is determined, and an initial power allocation demand vector is generated accordingly. This vector contains the target power direction and amplitude range for each energy unit in the next cycle. In other optional implementations, the prediction model can employ different types of time series prediction algorithms, such as recurrent neural networks, support vector regression, or autoregressive integral moving average models, to adapt to different climate and electricity consumption patterns.

[0038] Step S300: Perform global optimization to generate the power flow matrix.

[0039] To minimize system energy transmission losses and battery life losses, a multi-particle path search strategy is employed to optimize the initial allocation demand vector. Each particle represents a possible energy flow configuration scheme. The algorithm evaluates the comprehensive cost of each scheme during iteration and guides the population towards the optimal solution. The final output is an energy flow matrix that satisfies the system constraints, where each element represents the planned power value from a specific energy output to a receiver. In other alternative implementations, the multi-particle path search strategy can be replaced with other heuristic optimization algorithms, such as genetic algorithms, simulated annealing, or ant colony optimization, adapting to different computational resource conditions while ensuring solution quality.

[0040] Step S400: Verify the battery discharge rate and perform local repair to generate a corrected power flow matrix.

[0041] The discharge power of battery cells involved in the power flow matrix is ​​subjected to safety verification to determine whether it exceeds the dynamic safe power boundary determined by the current health status of the battery. If the limit is not exceeded, the original matrix is ​​retained; if the limit is exceeded, other legal paths in the matrix are locked unchanged, and the repair mechanism is triggered only for the path that exceeds the limit. This mechanism introduces a linear programming algorithm to redistribute the power difference caused by the reduction among available backup energy nodes, thereby completing a fast correction without interrupting the overall scheduling and obtaining the corrected power flow matrix. In some other optional implementations, the linear programming algorithm can be combined with the interior-point method or the dual simplex method to improve the computational efficiency under complex topologies.

[0042] Step S500: Convert the corrected power flow matrix into control commands and drive the hardware to execute them.

[0043] The corrected power flow matrix is ​​parsed into specific current or power control commands, which are then sent to the photovoltaic inverter, energy storage bidirectional converter, and charging pile controller, respectively. Each terminal device adjusts its own operating state according to the received commands to realize the actual flow and distribution of electrical energy. In some other optional implementations, the control command conversion process may include a protocol adaptation layer to ensure compatibility with industrial communication standards such as Modbus, CAN, or IEC 61850, ensuring interoperability with equipment from different manufacturers.

[0044] The aforementioned technical features form a close synergy: First, a comprehensive set of environmental variables is constructed through multi-source data fusion, providing comprehensive input for subsequent predictions; the prediction results drive the initial allocation logic, forming a preliminary scheduling intention; based on this, the global optimization algorithm explores the balance between economy and lifespan loss, generating a refined power flow scheme; however, due to the time-varying and nonlinear nature of the battery's physical state, directly executing the global optimal solution may pose safety risks. Therefore, a dynamic boundary verification mechanism based on health indicators is introduced to intercept high-risk operations in advance; when a potential limit violation is detected, the existing plan is not overturned, but rather quickly adjusted through a local repair strategy, ensuring both real-time performance and system stability; finally, the corrected instructions are parsed at the lower level and drive the power devices to operate, completing the closed loop from decision-making to execution. This layered architecture of "global planning—local error correction—precise execution" effectively solves the problem in existing technologies where safety and economy are difficult to balance due to insufficient battery aging perception, delayed optimization algorithm response, and lack of emergency repair capabilities.

[0045] Compared with the prior art, this embodiment, through the above-mentioned solution, can actively avoid the overload operation risk of aging batteries while ensuring the overall energy utilization efficiency of the system; it adopts a local repair mechanism to replace global recalculation, which significantly improves the response speed and fault tolerance of the control system; it realizes the transformation from static scheduling to dynamic adaptive energy management, and enhances the operational reliability and life-cycle economy of photovoltaic energy storage charging stations under complex working conditions.

[0046] In one specific implementation, the multi-dimensional sensing and acquisition module, through a sensor network deployed in the photovoltaic array, charging pile, and energy storage battery cluster, first synchronously acquires real-time irradiance and backsheet temperature data of the photovoltaic panels, simultaneously acquires the access voltage of the charging pile port and the requested current reported by the vehicle's BMS, and reads the terminal voltage, real-time temperature, and historically accumulated state of health (SOH) estimates of each individual battery cell from the energy storage battery management system (BMS). Specifically, the system uses principal component analysis (PCA) algorithm to extract features from the aforementioned raw high-dimensional data, and by calculating the eigenvectors of the covariance matrix, it extracts the irradiance... With backplate temperature The coupling effect is projected onto the direction of the first principal component, generating a supply intensity factor characterizing the photovoltaic output potential. Meanwhile, the voltage fluctuation slope, temperature rise rate, and static internal resistance of a single cell are compared. The resulting multivariate combination is projected onto the direction of the second principal component to extract orthogonal health decline factors. This factor can effectively reflect the increasing trend of polarization effect caused by battery electrochemical aging. Then, , Together with the real-time measured photovoltaic power generation and charging load power, they are encapsulated into a structured data package, forming a comprehensive set of environmental variables. This serves as the input vector for subsequent prediction models. In other alternative implementations, the feature extraction algorithm can be replaced with Linear Discriminant Analysis (LDA) or an autoencoder neural network to achieve nonlinear dimensionality reduction while preserving physical meaning; the construction of the health degradation factor can also incorporate fusion parameters of battery cycle count and calendar aging model to enhance the ability to characterize long-term degradation trends.

[0047] Through the above-described scheme, this embodiment can transform dispersed and heterogeneous field sensor data into low-dimensional feature vectors with clear physical meaning, significantly reducing the input dimension of subsequent prediction and optimization models and improving computational efficiency. Simultaneously, by separating external environmental disturbances from internal aging factors, the control strategy can respond specifically to changes in the battery's actual health state, avoiding the risk of overload on aging batteries due to a single SOC index. Compared with existing technologies, this embodiment enhances the observability of the battery's electrochemical degradation process and provides quantifiable state data for dynamic power boundary and variable gain control, thereby improving the precision control level of the energy management system without increasing hardware costs.

[0048] In one specific implementation, the process of generating the initial allocation demand vector includes: calculating the difference between the photovoltaic supply forecast and the charging demand forecast to determine whether a power gap exists. Specifically, when the difference is less than zero, it indicates that photovoltaic power generation is insufficient to meet the charging load demand, and the system enters battery discharge mode. Then, the remaining capacity of the energy storage battery in this state is assessed. If the battery's SOC (State of Charge) is detected to be higher than a preset minimum maintenance capacity threshold, such as 20% or 30%, it is determined that the battery has the capability to participate in power compensation. Next, the required discharge power gap to be borne by the energy storage battery is determined based on the absolute value of the supply-demand difference, and this power value is written into the initial allocation demand vector as a hard constraint for energy balance calculations in subsequent optimization processes.

[0049] In other alternative implementations, the minimum sustaining capacity threshold can be dynamically adjusted based on the battery type: for lithium iron phosphate batteries, this threshold can be set to 20%, while for ternary lithium batteries, it can be set to 15%; alternatively, the threshold can also be adjusted based on ambient temperature, automatically increasing the threshold in low-temperature environments to prevent the risk of lithium plating caused by deep discharge. Furthermore, the allocation of the discharge power gap can be further subdivided into a segmented release strategy across multiple time steps to avoid the impact of instantaneous high current output on battery life.

[0050] Through the above solution, this embodiment effectively avoids the risk of over-discharge caused by blindly using low-charge batteries for compensation when power supply is insufficient, ensuring that the energy storage unit always operates within a safe operating range. Compared with the prior art, this embodiment introduces a SOC-based state criterion during the initial allocation demand vector generation stage, realizing the pre-processing of battery physical constraints. This not only improves the safety of the system but also reduces the frequency of triggering subsequent correction mechanisms due to exceeding limits, thereby reducing the iterative overhead of the control algorithm and improving the overall scheduling efficiency.

[0051] In one specific implementation, the execution of the multi-particle path search strategy first includes initializing the particle swarm. Specifically, each particle is assigned a position vector that encodes a feasible energy allocation scheme, representing the power flow direction that covers the interaction between multiple energy sources and load nodes in the system. The particle's velocity vector is used to drive its iterative optimization in the solution space. The fitness function is constructed as a multi-objective optimization expression containing at least three weighted components: the first reflects the ohmic loss level in the energy transfer process within the system, the second quantifies the economic cost of energy interaction with the external power grid, and the third is a nonlinear penalty term related to battery temperature rise. In each iteration, for the position scheme corresponding to the current particle, the system estimates the Joule heating effect based on the product of the predicted current value and the battery's dynamic internal resistance, and derives the corresponding expected battery temperature rise value by combining the thermal time constant model. If the expected temperature rise value exceeds a preset safety threshold, the particle's position is marked as an invalid solution in the infeasible region, and its fitness value is forcibly degraded, thereby guiding the particle swarm's search direction to avoid such high-risk operating ranges. Through continuous updates of individual and global extrema, the entire population gradually converges to a set of Pareto approximate optimal solutions that take into account energy efficiency, economy, and thermal safety.

[0052] In other alternative implementations, the multi-particle path search strategy can be replaced by a genetic algorithm or an ant colony optimization algorithm, where the population individuals also encode the power allocation logic in matrix form; the penalty term in the fitness function can also be implemented using a piecewise linear function or a lookup table-based mapping relationship, rather than a continuous mathematical expression; the calculation of the expected temperature rise can also incorporate the environmental heat dissipation coefficient or battery package thermal resistance parameter to improve prediction accuracy.

[0053] Through the above scheme, this embodiment can effectively avoid the problem of generating aggressive scheduling plans when the traditional global optimization method ignores the battery thermal safety constraints, realize intelligent optimization of energy flow under complex multi-objective conditions, and incorporate the risk of thermal runaway caused by electrochemical aging into the decision-making considerations, which significantly improves the safety boundary and long-term stability of the system operation.

[0054] In one specific implementation, the power flow matrix is ​​implemented as a two-dimensional matrix structure, where each row corresponds to an energy output source, including photovoltaic systems, energy storage batteries, and the power grid; each column corresponds to an energy receiver, namely energy storage batteries, charging piles, and grid feedback channels. Each element in the matrix represents the real-time power value transferred from the energy side represented by the corresponding row to the load or storage side represented by the corresponding column, in kilowatts (kW). For example, the element located in the 1st row and 2nd column... This indicates the active power directly supplied to the charging pile by the photovoltaic power generation unit.

[0055] Specifically, when generating this two-dimensional matrix, the global optimization process first obtains the actual power demand of each receiver in the current scheduling period, such as the total load of the charging pile group. Charging targets for energy storage batteries And the maximum power limit allowed to be fed back to the grid. Then, during particle swarm optimization, for each candidate solution, i.e., the power flow matrix corresponding to each particle path, a column conservation check is performed: for each energy receiver column vector, the sum of the power contributions from all its input sources is calculated, and it is ensured that the sum equals the net power demand of the receiver at the current moment. Taking the column containing the charging pile as an example, it must satisfy: in, (Direct photovoltaic power supply component): This refers to the active power directly supplied to the charging pile by the photovoltaic power generation unit. The system prioritizes maximizing this component to improve the clean energy consumption rate. (Energy Storage Valley Filling Component): This represents the active power released from the energy storage battery unit to the charging pile. This parameter serves as a flexible adjustment variable for the system. When photovoltaic power generation is insufficient or during periods of high electricity prices, the algorithm will increase this value to fill the power gap, achieving the effects of "peak shaving and valley filling" and smoothing out fluctuations. (Grid backup component): This represents the active power directly extracted from the external grid and supplied to the charging piles. This parameter serves as a backup power variable for the system, in case the total output of distributed energy resources (photovoltaics and energy storage) cannot meet the total load demand. When the battery is in a low-power protection state, the algorithm will dynamically adjust this value to ensure that the user's charging service is not interrupted.

[0056] If the above equation is not satisfied, the particle's position is considered to violate physical constraints, and its fitness value is penalized to guide it towards convergence to the feasible region. Through the dynamic coupling and complementarity of the above three components, this embodiment can flexibly find the economically optimal energy ratio while meeting rigid charging requirements.

[0057] Next, the power balance state of all columns is continuously monitored during the optimization iteration. The matrix is ​​only considered a valid feasible solution when all columns satisfy the energy conservation condition and other boundary constraints (such as device capacity and dynamic safety boundaries) are also satisfied. This constraint mechanism is implemented by building a hard equality constraint submodule within the algorithm, which is integrated into the optimization solver to evaluate the constraint satisfaction in real time after each particle update.

[0058] In some other alternative implementations, the dimensions of the two-dimensional matrix can be extended according to the actual system topology, for example, by adding a backup diesel generator as the fourth row / column, or by splitting the charging pile into two independent load nodes: fast charging and slow charging. In addition, the energy conservation constraint can also be in the form of a soft constraint, which is approximately satisfied by introducing a weighted deviation term in the fitness function, and is suitable for actual working conditions with measurement errors or communication delays.

[0059] Through the above scheme, this embodiment ensures that the generated power flow matrix strictly follows the law of energy conservation, avoiding system instability caused by power allocation redundancy or gaps. Compared with the prior art, this embodiment not only achieves accurate power matching between multiple sources and loads, but also significantly improves the engineering feasibility and real-time execution reliability of the scheduling scheme by embedding physical laws into the optimization search process, making it particularly suitable for the long-term stable operation of microgrid systems with complex topologies.

[0060] In one specific implementation, the generation of the dynamic safe power boundary is first modeled based on the nonlinear mapping relationship between battery health index data and the maximum permissible charge / discharge rate. Specifically, the system pre-stores a correlation function between the state of health (SOH) and the maximum permissible charge / discharge rate (C-rate). This function is a monotonically decreasing curve obtained through empirical fitting. For example, when the SOH is 100%, the maximum permissible discharge rate is 1.5C; when the SOH drops to 80%, the corresponding rate automatically shrinks to 0.9C; when the SOH further drops to 70%, the rate is limited to 0.6C. This mapping relationship outputs the basic rate limit in real time through a lookup table or analytical function.

[0061] Then, the reduced rate of operation is weighted and corrected based on the real-time temperature of the energy storage battery. The temperature correction coefficient β(T) is implemented using a piecewise linear function: when the battery operating temperature is between 25℃ and 35℃, β(T) = 1.0, with no adjustment; when the temperature rises to 40℃, β(T) linearly decreases to 0.85; when the temperature reaches 45℃, β(T) further decreases to 0.7 to suppress the risk of high-current operation at high temperatures; when the temperature exceeds 50℃, β(T) drops sharply to 0.3, achieving strong constraint protection. The final dynamic safe power boundary is calculated by the following formula: in This is an estimated value for the battery open-circuit voltage, obtained, for example, by looking up a table using the open-circuit voltage-state-of-charge (OCV-SOC) curve. This is the base rate corrected by SOH. This is the temperature weighting coefficient. This boundary value is updated once per second and serves as a hard constraint input to the subsequent power allocation verification process.

[0062] In some other alternative implementations, the nonlinear mapping relationship can be obtained through online training of a machine learning model, and the decay curve between SOH and C-rate can be continuously optimized using historical operating data; or, a thermal time constant can be introduced into the temperature correction coefficient, taking into account the influence of battery thermal inertia, and a first-order hysteresis filter can be used to process the real-time temperature sampling value to avoid frequent jumps in control parameters due to instantaneous temperature rise.

[0063] Through the above scheme, this embodiment can dynamically adjust the power output capability of the battery according to its aging degree and real-time operating conditions. It actively derating the battery when its internal resistance increases or its capacity decreases, preventing safety hazards such as overheating and voltage exceeding limits under high loads. Compared with existing technologies, this embodiment not only transforms the battery health status from a passively monitored parameter into a variable actively involved in control decisions, but also achieves adaptive evolution of the safety boundary with electrochemical characteristics through a multi-dimensional coupling correction mechanism, significantly improving the availability and operational reliability of aging battery packs under complex operating conditions.

[0064] In one specific implementation, the process of reallocating power differences among standby nodes with power adjustment margins within the system using a linear programming algorithm includes the following steps: When the planned battery discharge power in the power flow matrix exceeds the dynamic safety power boundary generated in real time based on health indicators, the difference between the planned power in the over-limit path and the dynamic safety power boundary is first calculated to obtain the power to be transferred. Specifically, if the planned power output by the global optimization module for the battery to supply power to the charging pile is 50kW, but the dynamic safety power boundary is currently limited to 35kW due to increased battery temperature and decreased SOH, a power gap of 15kW to be transferred is generated. Then, standby nodes with carrying capacity are identified in the power flow matrix, including other healthy battery clusters or grid connection nodes with load rates below 80% of rated capacity. The availability of grid connection nodes is determined based on their real-time electricity price status and grid-connected capacity. Next, a linear programming model is constructed with minimizing the total allocation deviation as the objective function. The variables are the power proportion undertaken by each standby node, and the constraints include the maximum allowable input / output power of each node and the system energy balance requirements. The model was solved using the simplex method to calculate the optimal solution whereby the entire 15kW power to be transferred is handled by the grid side, and corresponding correction commands were generated. In other optional implementations, the backup node may also include a hydrogen energy storage unit or a heat load regulating device, provided that these devices have dispatchable margins under the current operating conditions; the objective function may also incorporate an economic cost weight to prioritize the allocation of the power to be transferred to the node with the lowest unit regulation cost.

[0065] By employing the above approach, this embodiment avoids control failures or system tripping caused by exceeding battery safety boundaries, achieving rapid local repair of the power allocation scheme without re-executing the global optimization algorithm. This method significantly reduces response latency, ensures the real-time performance and security of scheduling commands, and maintains overall system operating efficiency through optimal allocation strategies.

[0066] In one specific implementation, after converting the corrected power flow matrix into current control commands and sending them to the energy storage bidirectional converter, a real-time smoothing operation is further performed to address second-level power fluctuations on the load side. Specifically, within a millisecond-level sampling period, the instantaneous power data of the actual load is continuously collected by the charging pile controller; this sampling frequency is no less than 100Hz to ensure that rapid power changes caused by vehicle BMS adjustments or user plugging and unplugging of the charging gun during electric vehicle charging can be captured.

[0067] The actual load power at the current moment is compared with the planned power supply capacity for the corresponding time period in the corrected power flow matrix, and the instantaneous deviation between the two is calculated.

[0068] A variable gain damping controller within a hardware execution control module generates a dynamic compensation signal based on the instantaneous deviation. This compensation signal is superimposed on the reference power command of the bidirectional energy storage converter for real-time adjustment of its output power. In this implementation, the variable gain damping controller in the hardware execution control module employs a PID control structure, with parameters preset based on the system's historical operational stability and possessing online adjustment capabilities. In other optional implementations, the variable gain damping controller in the hardware execution control module can be replaced with a proportional resonant (PR) controller, a fuzzy PID controller, or a model predictive control (MPC) algorithm to adapt to energy storage converter hardware platforms with different response characteristic requirements.

[0069] Through the above scheme, this embodiment can quickly respond to short-term power disturbances outside the master control command update cycle, avoiding bus voltage fluctuations or protection actions caused by instantaneous supply and demand mismatch; at the same time, by introducing a feedback compensation mechanism based on measured deviation, the execution accuracy and system robustness of the energy distribution strategy under dynamic operating conditions are improved.

[0070] In one specific implementation, the process of generating a compensation signal for the bidirectional converter of the energy storage using a PID controller, i.e., a variable gain damping controller in the hardware execution control module, includes performing adaptive gain adjustment based on battery state, including: acquiring the dynamic impedance parameters of the energy storage battery in real time through the BMS. This parameter is determined by the controller based on the change in battery terminal voltage during the sampling period. With output current change The ratio is calculated online, that is... Specifically, when detected Exceeding the preset aging benchmark value (For example, when the nominal internal resistance of a new battery is 10mΩ), the controller automatically adjusts the proportional gain coefficient in the PID control loop. and differential gain coefficient This causes the proportional gain to decrease non-linearly with increasing impedance. The adjustment of the proportional gain follows the formula... ,in For the initial calibration gain, The damping sensitivity coefficient is used; the differential gain decays exponentially. , This is the decay rate factor. Then, the adjusted gain parameter is directly substituted into the PID calculation to generate compensation commands for the bidirectional converter of the energy storage system, thereby limiting the rate of change of current under high internal resistance conditions. Next, when subsequent monitoring shows that the battery dynamic impedance has fallen back to the normal range (e.g., close to...),... The controller gradually recovered. and To its nominal value, to maintain a fast response to load fluctuations. In some other alternative implementations, the dynamic impedance can also be estimated by fitting the slope of the voltage-current transient response curve using a sliding window least squares method; or the gain adjustment strategy can be replaced by a lookup table method based on fuzzy rules, according to... Use temperature as a dual input to query the preset gain correction coefficient matrix.

[0071] Through the above solution, this embodiment can effectively suppress the risk of instantaneous voltage drop caused by battery aging without sacrificing system response performance. Compared with the prior art, this embodiment achieves a differentiated control strategy of "slow response for old batteries and fast adjustment for new batteries" by dynamically coupling the gain parameter of the PID controller with the actual electrochemical state of the battery. This avoids the problem of aging batteries easily triggering undervoltage protection under fixed parameter control, and significantly improves the operational stability and safety of multi-lifecycle energy storage units under power surge conditions.

[0072] In one specific implementation, the real-time smoothing step further includes dead-zone control logic: setting a response dead-zone threshold for power fluctuations, which is used to determine whether to activate the PID controller to compensate for the bidirectional converter in the energy storage system. Specifically, this dead-zone threshold is configured as a fixed power range, such as ±200W, and its value can be adjusted proportionally according to the system's rated power. In charging stations of different capacity levels, it can be set between 0.5% and 1% of the rated power. When the absolute value of the instantaneous deviation between the actual load and the planned power is less than or equal to this threshold, the system determines that the fluctuation is within an acceptable range, does not trigger any control action, and maintains the current output power of the bidirectional converter in place. At this time, the minor power imbalance is absorbed by the grid side through natural power flow without active intervention, thereby avoiding the losses and stress accumulation caused by frequent start-stop battery charging and discharging.

[0073] The PID controller is activated to generate a corresponding compensation signal only when the absolute value of the instantaneous deviation exceeds a preset dead zone threshold. This signal is then superimposed on the reference power command of the bidirectional energy storage converter to dynamically correct energy distribution. This mechanism effectively filters out high-frequency, low-amplitude disturbances caused by minor adjustments in electric vehicle charging current and slight fluctuations in photovoltaic output, significantly reducing the operating frequency of the control system. In other optional implementations, the dead zone threshold can be set as an adaptive variable, dynamically adjusted based on the battery state of health (SOH) or ambient temperature. For example, when the battery SOH is below 80%, the dead zone range can be appropriately expanded to further reduce the adjustment burden on aging batteries; or, during historical periods of stable load changes, the threshold can be automatically relaxed based on a predictive model.

[0074] Through the above solution, this embodiment can effectively suppress over-response to minute power fluctuations, reduce unnecessary cycle counts of energy storage devices, and extend battery life; at the same time, it reduces the occupation of computing and communication resources in the control system, improving overall operating efficiency. Compared with the prior art, this embodiment achieves refined management of control actions while ensuring power supply continuity and voltage stability, avoiding energy waste and device wear caused by "control for control's sake".

[0075] In one specific implementation, the method further includes a long-term evolution and parameter update mechanism, comprising: daily packaging and storing the comprehensive set of environmental variables generated during operation, the corrected power flow matrix, and the deviation data between the actual execution results and the plan, constructing a structured historical scenario log, and archiving it by timestamp in a local or industrial big data center, i.e., a cloud database. Specifically, each log record contains an environmental feature vector for that day, the dimensions of which cover quantifiable indicators such as daily cumulative irradiance, average temperature, peak load power, and average battery pack health status.

[0076] Before the start of a new scheduling cycle, the system extracts the corresponding environmental feature vector for the current prediction day and calculates its similarity with each record in the historical log. This process uses cosine similarity in the vector space model as a metric, evaluating the degree of closeness between the current feature vector and each historical vector in the multidimensional feature space by calculating the cosine value of the angle between them.

[0077] The most similar historical records are identified as reference scenarios, and control correction parameters matching these scenarios are extracted. These correction parameters are injected as external feedback signals into the control flow of the current cycle to dynamically adjust the constraint boundaries of the energy allocation strategy or optimize the target weights.

[0078] In other alternative implementations, the similarity matching algorithm can be replaced with Euclidean distance, Pearson correlation coefficient, or machine learning-based clustering models (such as K-Means); the historical logs can be stored in a time-series database, a relational database, or a distributed file system to suit the data management needs of sites of different sizes.

[0079] Through the above scheme, this embodiment enables the control system to memorize and reuse historical operating experience, allowing the energy allocation strategy to adaptively optimize as the operating environment evolves. Compared with existing technologies, this embodiment avoids the repeated trial-and-error process of control logic, improves the consistency and robustness of system decisions under similar weather and load conditions, and provides a data-driven technical path for personalized strategy optimization in complex microgrid scenarios.

[0080] In one specific implementation, the process of extracting and applying correction parameters under the reference scenario includes: first, extracting the operating data curves of the battery cells in the historical day, specifically including the temperature rise curve and terminal voltage fluctuation curve of the battery cluster, and performing time-series segmentation analysis on the curves. Specifically, the system divides the historical day into multiple time windows (e.g., each 15 minutes is a time period), and for each window, it determines whether there is a situation where the battery temperature exceeds 42°C for more than 5 minutes. If so, it determines that an overheating event has occurred in that time period. Similarly, it detects whether there is a situation where the terminal voltage drops instantaneously by more than 8% of the rated voltage. If so, it determines that an overvoltage / undervoltage risk event has occurred. Then, the system generates corresponding correction instructions: for the same or similar time period of the current predicted day, the dynamic safety power boundary module automatically lowers the upper limit of the power for that time period by 10% to 15% to avoid the risk of heat accumulation in advance. Next, the system further analyzes whether solar curtailment occurred on historical days, i.e., solar power generation exceeds the sum of load and energy storage absorption capacity, and the duration exceeds 10 minutes. If curtailment is detected, the global path optimization module adjusts the weight coefficients in the fitness function, increasing the reward weight of the solar power consumption item. Increasing the weight by 0.1 to 0.3 units allows for prioritizing power allocation paths that can absorb more photovoltaic power locally during particle swarm optimization. In other alternative implementations, the threshold for determining temperature rise events can be differentiated based on different battery chemistry systems (such as lithium iron phosphate or ternary materials); the weight adjustment method can also use a nonlinear mapping table instead of a fixed increment, or dynamically output the optimal weight combination through a reinforcement learning model.

[0081] Through the above scheme, this embodiment can achieve feedforward parameter tuning based on historical failure modes, enabling the control system to proactively adopt a more conservative safety strategy or a more aggressive energy recovery strategy when facing similar operating conditions, thereby avoiding the recurrence of battery thermal runaway or energy waste. Compared with the prior art, this embodiment not only achieves automatic iterative optimization of control parameters, but also improves the robustness and long-term energy efficiency of the system in complex operating environments by combining physical state anomaly recording with economic target adjustment.

[0082] In one specific implementation, the battery health index data includes not only the real-time state of health (SOH) value, but also a dynamic aging characteristic: the internal resistance growth rate. The system periodically collects voltage, current, and temperature data of individual cells through the BMS, and estimates the current battery internal resistance value using the open-circuit voltage method and pulse discharge testing. Specifically, a baseline internal resistance sampling is performed every 24 hours, and the rate of change of internal resistance ΔR / Δt between two adjacent sampling periods is calculated. When this rate of change exceeds a preset aging rate threshold (e.g., an average daily increase greater than 0.8% for three consecutive days), the battery cell is determined to have entered the accelerated aging stage.

[0083] During the verification of the power flow matrix, if the internal resistance growth rate of a certain battery cluster is found to exceed the standard, the charging power limiting logic is immediately triggered: the charging power originally allocated to that battery cluster is forcibly reduced to 40% of the rated maximum charging power, and the energy storage task is preferentially transferred to other battery clusters in good health. This limitation applies to all charging scenarios, regardless of whether the photovoltaic surplus is sufficient, to prevent high internal resistance batteries from experiencing a rapid rise in terminal voltage to the cutoff voltage due to ohmic heat accumulation during charging, resulting in a "false high voltage" phenomenon, where the actual capacity is not full but the voltage has reached the upper limit, forcing the termination of charging, thereby affecting the overall energy storage efficiency and the usable battery capacity.

[0084] The system transmits the power redistribution demand to the safety boundary correction module, which calls the linear programming solver to redistribute the energy to be stored among the remaining healthy energy storage units, ensuring that the total energy storage target remains unchanged while avoiding the overload risk of aging batteries. In some other optional implementations, the monitoring of the internal resistance growth rate can be smoothed by sliding window averaging filtering or exponentially weighted moving average (EWMA) to eliminate misjudgments caused by instantaneous measurement noise; alternatively, a trend prediction model based on machine learning can be used to predict the internal resistance trend in the next few days and pre-activate the current limiting mechanism before the trend reaches a threshold.

[0085] Through the above scheme, this embodiment can effectively identify energy storage battery cells in the accelerated aging stage and actively apply charging constraints in the early stage of significant degradation of their electrochemical performance, so as to avoid abnormal rise in terminal voltage caused by excessive internal resistance and the resulting problems such as premature charging stop and reduced capacity utilization, thereby improving the overall coordination and control capability and long-term operation stability of multi-cluster parallel energy storage system.

[0086] In one specific implementation, this dynamic energy allocation optimization method is applied to a photovoltaic energy storage charging station system comprising multiple parallel energy storage battery clusters. In practice, the central control unit establishes a dedicated energy flow planning channel for each independent energy storage battery cluster, so that the power flow matrix is ​​no longer based on a single battery system, but rather generates independent charging and discharging power commands for each battery cluster. Specifically, in the global optimization phase, the particle swarm optimization algorithm individually evaluates the SOH, real-time temperature, and internal resistance data of each battery cluster and embeds them as constraints into their respective fitness functions, thereby outputting differentiated discharge plans. In the safety boundary correction phase, the calculation of the dynamic safety power boundary is based on the health indicators of each battery cluster. This results in the maximum allowable charging and discharging power of battery clusters with poor health (e.g., SOH below 80% or internal resistance above the average by 15%) being actively compressed; while battery clusters with health conditions better than the average retain higher power response capabilities. The revised power flow matrix achieves load rebalancing through a weighted allocation mechanism: when the system needs to discharge, high-healthy battery clusters are prioritized to undertake additional power tasks; when charging is required, the input current of low-healthy battery clusters is limited to prevent them from reaching the cutoff threshold prematurely due to artificially high terminal voltage. In other optional implementations, task allocation among battery clusters can be achieved by introducing a priority queue mechanism, where the scheduling priority is reduced by one level for every 5% decrease in health indicators; or a linear allocation strategy based on weight factors can be adopted, mapping the C-rate capability of each battery cluster proportionally to its SOH value, thereby achieving a continuous and smooth power allocation gradient.

[0087] Through the above solution, this embodiment enables differentiated and refined energy management of multi-battery cluster systems, effectively mitigating the overload risk of aging battery cells while ensuring overall power supply capacity. Compared with existing technologies, energy storage units in different health states can undertake matching power loads based on their actual electrochemical characteristics, significantly reducing the probability of thermal runaway or accelerated cycle life degradation in degraded batteries, while improving the utilization rate of the available capacity and long-term operational reliability of the entire energy storage system.

[0088] In one specific implementation, the sensor network continuously collects on-site operational data at a frequency higher than the generation cycle of the power flow matrix. Specifically, the photovoltaic inverter, charging pile controller, and battery management system (BMS) report real-time raw measurements such as power, voltage, current, and temperature at 1-second intervals, while the global energy allocation matrix is ​​generated every 15 minutes by a multi-particle path search strategy.

[0089] At the start of each data acquisition cycle, the latest acquired multidimensional sensing data is input into the supply and demand forecasting model. This model employs a sliding time window mechanism, retaining the historical data sequence of the most recent 30 minutes, and dynamically updating the forecast results for photovoltaic power generation and charging load demand in the next hour based on this. The system continuously compares the deviation between the actual observed value at the current moment and the predicted value in the previous round. When the relative error of a certain variable (such as photovoltaic output or total load) exceeds the preset replanning threshold (e.g., ±15%), the multi-particle path search strategy is immediately triggered to re-execute the process, generating a new power flow matrix in advance, without waiting for the original 15-minute cycle to end.

[0090] In the new round of optimization, the original initial population can be initialized based on the previous convergence results to speed up the solution process; at the same time, the corrected prediction vector output by the prediction model is used as input conditions to participate in the optimization calculation.

[0091] In some other alternative implementations, the duration of the sliding window can be adaptively adjusted according to the characteristics of seasonal light variation, for example, set to 20 minutes during periods of severe light fluctuations in spring and autumn, and extended to 40 minutes during periods of stable power output in winter; or, the replanning threshold can be dynamically adjusted according to the battery health status (SOH) - when the SOH is below 80%, the trigger threshold is reduced to ±10% due to the decrease in system fault tolerance, so as to improve the scheduling response sensitivity.

[0092] Through the above scheme, this embodiment can achieve rapid response to environmental changes and load disturbances, avoiding energy allocation failure due to inaccurate predictions; at the same time, it uses high-frequency sampling and sliding modeling mechanisms to enhance prediction continuity, ensuring that the system can maintain a balance between economy and safety under unplanned operating conditions, and improving the robustness and adaptability of the overall control strategy.

[0093] As shown in Figure 2, this embodiment of the invention provides a dynamic energy allocation optimization system for photovoltaic energy storage charging stations, including a multi-dimensional sensing and acquisition module M100, deployed at the photovoltaic energy storage charging station site. This module is configured to synchronously acquire real-time output data of photovoltaic power generation units, load demand information from charging piles, and BMS status parameters of energy storage battery packs via a distributed sensor network. These data are characterized by large volume, diverse types, and rapid generation, representing typical industrial big data in the energy storage field. The system then integrates and processes this multi-source heterogeneous data, mapping it into a unified set of comprehensive environmental variables. In one embodiment, the data collected by the multi-dimensional sensing and acquisition module M100 includes, but is not limited to, irradiance, photovoltaic module backsheet temperature, charging pile requested current, individual cell voltage, average battery cluster temperature, and estimated state of health (SOH). The module constructs a high-dimensional input vector by performing timestamp alignment and normalization on the original sensor data. Principal component analysis is then used to extract a supply intensity factor characterizing the volatility of external energy supply and a health degradation factor reflecting the aging trend of internal energy storage units. These two factors, along with the original power data, constitute a comprehensive set of environmental variables. In some other alternative implementations, the feature extraction process can be replaced by Independent Component Analysis (ICA) or an autoencoder network to achieve dimensionality reduction and semantic separation.

[0094] The supply and demand forecasting planning module M200 establishes a communication connection with the multi-dimensional sensing and acquisition module M100 and is configured to run a built-in trend forecasting model. Based on a comprehensive set of environmental variables, it generates forecasts of photovoltaic supply and charging load demand for future periods and generates an initial allocation demand vector according to the supply-demand gap and battery capacity constraints. In one embodiment, the supply and demand forecasting planning module M200 uses a Long Short-Term Memory (LSTM) neural network as its forecasting engine, with training data derived from historical operating records, covering typical weather patterns and electricity consumption cycles. The module calculates the net power deficit based on the forecast results and, combined with the battery SOC threshold, determines whether to initiate a discharge response. If the discharge conditions are met, it is written into the battery support item in the initial allocation demand vector. In other optional implementations, the forecasting model can use a gated recurrent unit (GRU) or a Transformer architecture instead of LSTM to accommodate the modeling needs of longer sequence dependencies.

[0095] The global path optimization module M300 is connected to the supply and demand forecasting and planning module M200 and is configured to execute the particle swarm optimization algorithm. Under the premise of satisfying energy conservation constraints, it uses minimizing overall system loss as the objective function to perform a global search for power flow paths between multiple energy nodes, outputting a power flow matrix containing the optimal energy allocation ratio among photovoltaics, batteries, and charging piles. In one embodiment, the global path optimization module M300 encodes the position of each particle as a... The system employs a two-dimensional matrix structure, where rows represent energy sources (photovoltaics, batteries, grid), columns represent energy destinations (battery charging, charging pile power supply, power return to the grid), and matrix elements represent the power values ​​along the corresponding paths. The fitness function comprehensively considers line transmission losses, grid interaction costs, and a battery life loss penalty based on temperature rise prediction, guiding the population to converge to an operation point that balances economy and safety. During the search process, the system enforces a demand balance constraint where the sum of the columns equals the target load. In other alternative implementations, the multi-objective optimization algorithm can be replaced with a genetic algorithm (GA) or ant colony optimization (ACO), and a Pareto front screening mechanism can be introduced to achieve multiple solution set outputs.

[0096] The safety boundary correction module M400 is connected to the output of the global path optimization module M300. It is configured to monitor in real time whether the power commands of each branch in the power flow matrix exceed the safe operating limits dynamically determined by battery health indicators. When an over-limit situation is detected, the unaffected energy path is locked, and the power gap is redistributed among available backup nodes using a linear programming method, generating a corrected power flow matrix that conforms to physical safety constraints. In one embodiment, after receiving the power flow matrix, the safety boundary correction module M400 first extracts the power commands related to battery charging and discharging, and then calls the dynamically generated safe power boundary based on battery health indicators for upper and lower limit verification. Once it is found that the power of a certain path exceeds the allowable range, the remaining non-over-limit paths remain unchanged, and only the abnormal branch is locally adjusted. Specifically, the system identifies backup resource nodes with adjustment capabilities (such as another battery cluster or grid interface), constructs a very small-scale linear programming problem, and aims to minimize the total deviation or cost increment caused by redistribution. The solver uses the simplex method to complete the calculation in milliseconds and output the correction scheme. In some other alternative implementations, the local repair strategy can be solved quickly using the interior point method or gradient projection method, which is particularly suitable for complex microgrid structures with higher dimensions.

[0097] The hardware execution control module M500, coupled with the safety boundary correction module M400, is configured to parse the corrected power flow matrix into underlying drive signals suitable for different power devices, and send them to the photovoltaic inverter, the energy storage bidirectional converter (PCS), and the charging pile controller, respectively, to achieve precise scheduling and execution of energy flow. In one embodiment, the hardware execution control module M500 receives the corrected power flow matrix, decomposes it into the set power values ​​of each terminal device, and converts them into standardized communication protocol frames (such as Modbus TCP or CANopen) and sends them to the corresponding controller. The photovoltaic inverter adjusts its MPPT operating point and output power limit accordingly, the energy storage PCS executes charge / discharge switching and power point tracking according to instructions, and the charging pile controller dynamically adjusts the output voltage and current of the DC / DC converter. The entire control link supports a second-level refresh rate to ensure accurate implementation of scheduling intentions. In other optional implementations, the drive signals can be transmitted in the digital substation architecture via IEC 61850 SV / GOOSE messages to improve real-time performance and interoperability.

[0098] The aforementioned modules form a progressive, closed-loop collaborative technology chain: the multi-dimensional sensing and acquisition module M100 provides physically meaningful state representations, laying the data foundation for subsequent decision-making; the supply and demand forecasting and planning module M200 generates a forward-looking scheduling blueprint based on this; the global path optimization module M300 seeks the optimal energy flow path on a macro scale, ensuring the economic efficiency of system operation; when the plan faces safety risks due to changes in battery health, the safety boundary correction module M400 does not overturn the original plan, but implements "minimally invasive" local repairs to avoid control delays caused by replanning; finally, the hardware execution control module M500 translates high-level instructions into low-level actions, completing the closed loop from logical decision-making to physical execution. This layered architecture effectively solves the contradiction between the time consumption of global optimization and real-time safety response in traditional control systems. Simultaneously, by introducing a health indicator-driven safety boundary mechanism, it achieves proactive protection of aging energy storage units, thereby improving the system's reliability and sustainable operation capabilities.

[0099] Through the above-described scheme, this embodiment can achieve efficient, safe, and stable energy distribution of photovoltaic energy storage charging stations in complex dynamic environments. Compared with existing technologies, this embodiment, by constructing a complete control link from environmental perception to command execution, not only achieves accurate prediction and global optimization under multi-source data fusion, but also establishes a dynamic safety verification mechanism based on battery health status. This prevents high internal resistance or low SOH batteries from being forced to participate in high-power scheduling, significantly reducing the risk of overload tripping and thermal runaway. At the same time, the adoption of a local linear repair strategy to replace global re-optimization greatly shortens the emergency response time and ensures the continuity and real-time nature of scheduling commands. The overall architecture supports flexible expansion and can be adapted to microgrid application scenarios of different scales and topologies.

[0100] In one specific implementation, the safety boundary correction module M400 integrates a health-power mapping unit. This unit has built-in non-volatile memory to store a non-linear lookup table between the battery's state of health (SOH) and the maximum permissible charge / discharge rate (C-rate). This lookup table is obtained through laboratory accelerated aging experiments and covers the typical degradation range of SOH from 100% to 70%, with each 5% representing a data point. The corresponding maximum C-rate value exhibits an exponential decay trend. For example, when SOH is 100%, the maximum permissible discharge rate is 2.0C; when SOH drops to 80%, the corresponding maximum discharge rate automatically adjusts to 1.2C; and when SOH further decreases to 75%, this value drops to 0.8C. The health-power mapping unit receives the SOH value from the BMS in real time, calculates the corresponding upper limit of C-rate at the current moment using a lookup table method combined with linear interpolation, and converts it into a dynamic safe power boundary based on the battery's rated capacity. The boundary value is directly input as a hard constraint to the linear programming solver in the safety boundary correction module M400, participating in the power redistribution calculation for out-of-limit paths. In some other optional implementations, the nonlinear lookup table can be dynamically updated through an online learning mechanism, with data sources including battery temperature rise curves, voltage response hysteresis, and cumulative cycle counts recorded during historical operation; alternatively, the lookup table can be stored in the embedded controller as a piecewise function, replacing the complete lookup table structure to reduce memory usage.

[0101] Through the above scheme, this embodiment can achieve precise quantitative constraints on the charge and discharge capabilities of aging batteries, avoiding overly conservative or risky runaway problems caused by fixed power limits. The health-power mapping unit transforms the battery's electrochemical degradation state into calculable control parameters, enabling the dynamic safety boundary to be continuously adjusted with changes in state of health (SOH), thus improving the adaptability and safety of the energy allocation strategy. Compared with existing technologies, this embodiment, without increasing additional hardware costs, achieves deep coupling between battery health state and power scheduling commands through a software-level nonlinear mapping mechanism. This effectively prevents high-internal-resistance batteries from being overloaded during scheduling, thereby reducing the risk of thermal runaway and extending the overall lifespan of the energy storage system.

[0102] In one specific implementation, the variable gain damping controller embedded in the hardware execution control module M500 is configured to receive the terminal voltage and output current sampling signals of the energy storage battery in real time, and perform differential calculations on the voltage difference and current difference between adjacent time moments based on a sliding time window to obtain the real-time dynamic internal resistance of the battery cell. The controller further incorporates an aging baseline storage unit to save the nominal internal resistance of each battery model under factory conditions. When detected (in When the preset aging threshold is reached, the controller automatically adjusts the differential gain parameter in the PID closed-loop control loop. According to the exponential decay function The current command is adjusted downwards to suppress drastic changes caused by sudden changes in external load. The controller also integrates a compensation signal synthesis unit, whose input is connected to the output of the PID calculation module, and whose output is coupled to the current loop reference input of the energy storage bidirectional converter through a communication interface. This unit is used to superimpose the generated compensation amount into the main control command. In some other optional implementations, the dynamic internal resistance estimation method can be replaced by an online identification algorithm based on extended Kalman filtering (EKF). In other optional implementations, the gain adjustment strategy can use a piecewise linear mapping table instead of a continuous exponential function to reduce the computational overhead of the embedded implementation.

[0103] Through the above solution, this embodiment can effectively alleviate the instantaneous voltage drop problem of high internal resistance aging batteries under power command changes, and avoid triggering BMS undervoltage protection and causing power interruption; at the same time, by adaptively adjusting the intensity of PID derivative action, differentiated response control for batteries in different health states is achieved, ensuring system stability while maintaining the necessary dynamic response capability to load fluctuations.

[0104] In one specific implementation, the system further includes a self-evolving learning engine, which establishes a data connection with the multi-dimensional sensing acquisition module M100 and the global path optimization module M300. This engine is configured to periodically receive a comprehensive set of environmental variables within the current operating cycle and construct the environmental feature vector for the day based on this set. The self-evolving learning engine maintains a non-volatile scene database, where each historical record contains a historical date identifier, a corresponding environmental feature vector, and a set of control parameters corresponding to the power flow direction matrix that is ultimately executed and verified by the safety boundary correction module M400 for that day. The control parameter set includes at least the initial population distribution of the PSO algorithm, the weight coefficients of each item in the fitness function, and the threshold for the number of convergence iterations. Before the start of each daily scheduling cycle, the self-evolving learning engine calculates the cosine similarity between the current day's environmental feature vector and each historical record, and selects the historical record with the highest similarity as the matching object. When the similarity exceeds a preset matching threshold (e.g., 0.92), the engine extracts the control parameter set from the historical record and injects it into the global path optimization module M300 as the initial population initialization basis and fitness function weighting benchmark for this PSO optimization. In some other alternative implementations, the calculation of cosine similarity can be replaced by Mahalanobis distance or dynamic time warping (DTW) methods to adapt to load pattern matching with time offset characteristics; in other alternative implementations, the injection of the control parameter set can adopt a weighted fusion strategy, that is, the new initial population is generated by a linear combination of the current default parameters and the historical best parameters according to the similarity ratio, rather than a complete replacement.

[0105] Through the above scheme, this embodiment can use historical operating experience to guide the global optimization algorithm to quickly converge to a high-quality solution region, avoid blind exploration in a complex search space, and significantly shorten the number of iterations required for optimization. At the same time, by feeding back the effective control parameters that have been verified in practice to the current scheduling process, the robustness and economic performance of the system under similar operating conditions are improved, which is especially suitable for photovoltaic energy storage charging station scenarios with obvious seasonal climate patterns or strong periodicity of load behavior.

[0106] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A dynamic energy allocation optimization method for photovoltaic energy storage charging stations, characterized in that, include: A sensor network deployed at photovoltaic panels, charging piles, and energy storage batteries is used to collect real-time power generation, load demand, battery remaining capacity, and health indicator data. This data is then mapped to a comprehensive set of environmental variables. This set of environmental variables is input into a prediction model to obtain predicted photovoltaic supply and charging demand values ​​for the future time period. An initial allocation demand vector is generated based on these predicted values. With the goal of minimizing system energy transmission loss and battery life loss, a multi-particle path search strategy is used to globally optimize the initial allocation demand vector, generating a vector containing photovoltaic power generation units and energy storage. A power flow matrix representing the energy distribution ratio between the battery unit and the charging pile unit is generated. The planned battery discharge rate in the power flow matrix is ​​checked to see if it exceeds the dynamic safety power boundary generated in real-time based on the health indicator data. If the battery discharge rate exceeds the dynamic safety power boundary, the flow paths in the power flow matrix that have not exceeded the boundary are locked. For the paths that exceed the boundary, a linear programming algorithm is used to redistribute the power difference among spare nodes with power adjustment margins within the system, resulting in a corrected power flow matrix. The corrected power flow matrix is ​​then converted into current control commands, which are sent to the photovoltaic inverter, the energy storage bidirectional converter, and the charging pile controller to drive the power devices to perform energy distribution.

2. The method according to claim 1, characterized in that, The process of mapping the real-time power generation, the load demand power, the remaining battery capacity, and the health indicator data into a comprehensive set of environmental variables specifically includes: synchronously collecting the irradiance and backsheet temperature of the photovoltaic panel, the access voltage and requested current of the charging pile, and the individual terminal voltage and real-time temperature of the energy storage battery; using a feature extraction algorithm to extract the supply intensity factor characterizing the output capacity of the photovoltaic array and the health degradation factor characterizing the degree of electrochemical aging of the battery from the collected data; and combining the supply intensity factor, the health degradation factor, and the real-time collected power data to form the comprehensive set of environmental variables.

3. The method according to claim 1, characterized in that, The process of generating the initial allocation demand vector specifically includes: calculating the difference between the photovoltaic supply forecast and the charging demand forecast; when the difference shows that the supply is less than the demand, determining whether the remaining battery capacity is higher than the minimum maintenance capacity; if the remaining battery capacity is higher than the minimum maintenance capacity, determining the discharge power gap that needs to be provided by the energy storage battery based on the difference, and writing the discharge power gap as a constraint condition into the initial allocation demand vector.

4. The method according to claim 1, characterized in that, The process of globally optimizing the initial allocation demand vector using a multi-particle path search strategy specifically includes: initializing a particle swarm, where the position vector of each particle represents a possible power flow matrix; constructing a fitness function, which includes an energy transmission efficiency term, a grid interaction cost term, and a battery temperature penalty term; during the search process, calculating the expected battery temperature rise value corresponding to each particle in real time, marking the positions of particles whose expected battery temperature rise value exceeds a preset safety value as invalid solutions, and guiding the particle swarm to converge to the position where the fitness function value is optimal and the expected battery temperature rise value is within the safety range.

5. The method according to claim 4, characterized in that, The power flow matrix is ​​a two-dimensional matrix, where the rows of the two-dimensional matrix represent energy output sources, the columns of the two-dimensional matrix represent energy receiving ends, and the element values ​​in the two-dimensional matrix represent the amount of power transmitted. The global optimization process also includes satisfying the energy conservation constraint, which is that the sum of the elements in each column of the two-dimensional matrix is ​​equal to the current power demand of the corresponding energy receiver.

6. The method according to claim 1, characterized in that, The process of generating the dynamic safe power boundary includes: establishing a nonlinear mapping relationship between the health index data and the maximum allowable charge / discharge rate; when the health index data indicates an increase in battery internal resistance or capacity decay, proportionally shrinking the value of the maximum allowable charge / discharge rate; and weighting and correcting the shrunk rate in conjunction with the real-time temperature of the energy storage battery to obtain the dynamic safe power boundary at the current moment.

7. The method according to claim 1, characterized in that, After sending the current control command, the method further includes a real-time smoothing step: collecting actual load fluctuation data of the charging pile at millisecond intervals; calculating the instantaneous deviation between the actual load fluctuation data and the corrected power flow matrix; A PID controller is used to generate a compensation signal for the energy storage bidirectional converter based on the instantaneous deviation.

8. The method according to claim 7, characterized in that, The process of generating a compensation signal for the bidirectional energy storage converter using a PID controller includes performing adaptive gain adjustment based on battery state: acquiring the dynamic impedance parameters of the energy storage battery in real time; when the dynamic impedance parameters increase, reducing the proportional gain coefficient and derivative gain coefficient of the PID controller to slow down the rate of change of the output current of the bidirectional energy storage converter; when the dynamic impedance parameters are within the normal range, restoring the proportional gain coefficient and derivative gain coefficient of the PID controller to improve the response speed to load fluctuations.

9. The method according to claim 1, characterized in that, The method further includes long-term evolution and parameter update steps: storing the comprehensive environmental variable set generated daily, the corrected power flow matrix, and the deviation data after actual execution as historical scene logs; calculating the cosine similarity between the environmental features of the current date and the environmental features of the historical dates recorded in the historical scene logs; selecting the historical date with the highest similarity as a reference scene, and extracting the correction parameters under the reference scene.

10. A dynamic energy distribution optimization system for photovoltaic energy storage charging stations, characterized in that, include: A multi-dimensional sensing and acquisition module, deployed at the station site, is configured to synchronously acquire photovoltaic power generation, charging pile load demand, and BMS status data of energy storage batteries through a sensor network, and map the photovoltaic power generation, charging pile load demand, and BMS status data into a comprehensive set of environmental variables. A supply and demand forecasting and planning module, connected to the multi-dimensional sensing and acquisition module, is configured to run a pre-set forecasting model to generate trend forecasts of photovoltaic supply and charging demand, and produce an initial allocation demand vector accordingly. A global path optimization module, connected to the supply and demand forecasting and planning module, is configured to execute a multi-particle path search algorithm, minimizing total system loss as the convergence objective while satisfying energy conservation, and outputting a power flow matrix among photovoltaic, battery, and charging pile components. A safety boundary correction module, connected to the global path optimization module, is configured to verify in real time whether the power flow matrix exceeds the dynamic safety power boundary constructed based on battery health indicators. This module is configured to recalculate the power allocation ratio among unbounded backup nodes using a linear programming algorithm when a boundary breach is detected, generating a corrected power flow matrix. The hardware execution control module is connected to the safety boundary correction module and is configured to parse the corrected power flow matrix into underlying drive signals and distribute them to the photovoltaic inverter, energy storage bidirectional converter and charging pile control unit.

Images