An energy management method, system and electronic device for a light storage and charging integrated charging station

Abstract

The application discloses a kind of considering battery health state's light storage fills integrated charging station energy management method, system and electronic equipment, belong to energy storage battery state control technical field.The method includes: the real-time operation data of light storage fills integrated charging station is collected, including photovoltaic power, charging demand and energy storage battery state;Battery aging cost model based on rain flow counting method and Arrhenius theory is constructed, the battery life loss under different charge-discharge behaviors is quantified;Establish deep reinforcement learning environment, to minimize the weighted sum of power purchase cost, battery aging cost and system penalty cost as the goal design reward function;Deep reinforcement learning algorithm is used to train decision model, and the trained model is used to output the optimal continuous charge-discharge power instruction online.The application can meet the demand of electric vehicle supercharging while effectively inhibiting the rapid decline of the life of energy storage battery due to high-rate charge-discharge, and realizing the collaborative optimization of economic benefit and equipment life.

Description

Technical Field

[0001] This invention relates to the field of dynamic control of energy storage batteries, and specifically to an energy management method, system and electronic equipment for an integrated photovoltaic-energy storage-charging station that takes into account the battery's health status. Background Technology

[0002] With the explosive growth of the new energy vehicle industry, integrated photovoltaic-storage-charging stations are being widely used as an important infrastructure for realizing the local consumption of green energy and alleviating the pressure of grid capacity expansion. In the photovoltaic-storage-charging system, the energy storage system plays a key role as an "energy buffer," which must both smooth out the random fluctuations in photovoltaic power generation and reduce the impact of peak electric vehicle charging on the power grid.

[0003] However, with the widespread adoption of high-power supercharging technology, the operating conditions of charging stations have significantly deteriorated: on the one hand, the instantaneous power demand of supercharging piles is extremely high and highly random; on the other hand, to maintain power balance, energy storage batteries need to frequently switch between high-rate charging and discharging. Under these conditions, existing photovoltaic-storage-charging energy management strategies mainly suffer from the following shortcomings: (1) Traditional rule-based or model predictive control (MPC) strategies usually treat energy storage batteries as ideal energy containers, and the optimization objective only includes the cost of purchasing electricity, ignoring the economic losses caused by the accelerated degradation of battery cycle life under high-rate operating conditions. In practice, the phenomenon of "saving electricity costs and replacing batteries" often occurs, resulting in a cost inversion of the entire life cycle, and the overall system benefits actually decrease.

[0004] (2) Cyclic life assessment based on rainflow counting requires complete charge and discharge cycle data, which is a post-event statistical method and cannot quantify instantaneous aging costs in a single control decision. While the Arrhenius equation can describe the temperature-accelerated aging effect, its application requires the prediction of future temperature curves, which is fundamentally contradictory to the real-time decision-making requirements.

[0005] (3) Rule logic is difficult to handle the dual strong randomness of photovoltaic and supercharging loads; MPC needs to solve high-dimensional non-convex optimization problems online, which is computationally burdensome and difficult to meet the fast response requirements of supercharging impact; discretized reinforcement learning algorithms such as Q-learning are prone to power command step jumps due to the discrete action space, which aggravates the mechanical and thermal stress inside the battery.

[0006] Therefore, there is an urgent need for an energy management method that can monetize battery aging costs in real time and integrate them into decision-making objectives while meeting supercharging requirements, thereby achieving continuous power optimization control and solving the technical problem of rapid degradation of the health status of energy storage batteries. Summary of the Invention

[0007] The purpose of this invention is to provide an energy management method and system for integrated photovoltaic-energy storage-charging stations that considers battery health status, aiming to solve the problem in existing technologies that neglect the aging cost of energy storage batteries, leading to excessively rapid battery life degradation in high-power overcharging scenarios. This invention achieves synergistic optimization of system operating economy and battery health life through deep reinforcement learning algorithms.

[0008] To achieve the above objectives, the present invention provides an energy management method for an integrated photovoltaic-storage-charging station that considers the battery's health status, comprising the following steps: S1. Data Acquisition and Preprocessing: Real-time acquisition of the operating status data of the photovoltaic-storage-charging integrated charging station system. The operating status data includes at least the real-time photovoltaic power generation, electric vehicle charging demand power, the current state of charge (SOC) of the energy storage battery, the current state of health (SOH) of the energy storage battery, and the grid transaction price. The operating status data is then normalized. S2. Construct a battery aging cost quantification model: Based on the rainflow counting method and the Arrhenius semi-empirical model, establish a mapping relationship model between battery life loss and charge / discharge behavior. This model is a battery aging cost quantification model suitable for deep reinforcement learning. Input the charge / discharge power or rate into the model to calculate the instantaneous battery aging cost corresponding to a single action. S3. Constructing a deep reinforcement learning decision-making framework: Define a state space that contains the normalized running state data from step S1; Define the action space as the charging and discharging power commands for the energy storage battery within a continuous interval; Define a reward function, which is configured to maximize the overall system benefit. Its components include at least grid interaction cost, battery aging loss cost obtained based on the single-step aging cost calculation model, and system default penalty cost. S4. Model Training: The Deep Deterministic Policy Gradient (DDPG) algorithm is used to train the policy network offline. The policy network outputs continuous charging and discharging power actions based on the input state vector. The cumulative reward is maximized through an experience replay mechanism and network parameter update strategy. S5. Online Control and Safe Execution: The trained policy network is deployed to the central controller of the charging station. Based on the real-time collected operating status data, charging and discharging power commands are generated and safety boundary verification is performed. Finally, the verified power commands are sent to the energy storage converter (PCS) for execution.

[0009] Furthermore, in step S2, the process of constructing the battery aging cost quantification model includes: Rainflow counting is used to process historical or simulated battery state-of-charge time series data to identify closed hysteresis loops and extract the discharge depth, cycle mean, and duration characteristics of effective charge-discharge cycles. A semi-empirical Arrhenius model was used to establish the functional relationship between battery capacity decay rate and charge / discharge rate, depth of discharge, and temperature. Based on the above functional relationship, offline simulations were performed at different charge and discharge rates to obtain a nonlinear function of battery cycle life and charge and discharge rate, which is the battery aging cost quantification model.

[0010] Furthermore, the battery capacity decay rate The expression is:

[0011] in, This is the forward exponent factor, which is related to the charge / discharge rate; Activation energy; It is the ideal gas constant; This refers to the absolute temperature of the battery. Accumulate and process power for the battery; This is the power-law exponent.

[0012] Furthermore, in step S3, the reward function The expression is as follows:

[0013] in, This represents the cost of interaction with the power grid; The battery aging loss cost is represented by the battery aging cost quantification model obtained through step S2. This represents the system's default penalty cost, which includes at least the power shortage penalty for failing to meet the charging needs of electric vehicles and the penalty for exceeding the SOC limit of the energy storage battery. These are the corresponding weighting coefficients.

[0014] Furthermore, in step S4, the specific process of offline training using the Deep Deterministic Policy Gradient (DDPG) algorithm is as follows: (1) Initialize the experience replay pool, and initialize the current network of the Actor, the target network of the Actor, the current network of the Critic, and the target network of the Critic; (2) The Agent outputs the energy storage charging and discharging power action through the Actor network according to the current state, and adds random noise to the action to increase randomness; (3) After the system performs the action, the environment is transferred to the next state, and the immediate reward is calculated according to the reward function described in step S3. The quadruple consisting of state, action, reward and next state is stored in the experience replay pool. (4) When the amount of data in the experience replay pool reaches the preset threshold, a small batch of data is randomly sampled, the parameters of the Critic network are updated to minimize the temporal difference error, and the parameters of the Actor network are updated to maximize the Q value of the Critic network output. (5) Use a soft update strategy to slowly update the target network parameters until the cumulative reward value tends to stabilize.

[0015] Furthermore, in step S5, the security boundary verification specifically includes: If the current state of charge of the energy storage battery And the generated charge / discharge power command is in the charging direction ( Then, the power command will be forcibly modified. =0; If the current state of charge of the energy storage battery Furthermore, the generated charge / discharge power command is in the discharge direction. Then force correction of power command =0; If the absolute value of the generated charging / discharging power command Exceeding the converter's maximum allowable power Then, the amplitude will be limited to ± .

[0016] Furthermore, the power demand for electric vehicle charging includes the high-power random load of supercharging piles; during the training process of the DDPG algorithm, random noise is superimposed on the power load of the supercharging piles to simulate the uncertainty of the supercharging load and improve the algorithm's adaptability to high-rate charging and discharging conditions.

[0017] To achieve the above objectives, the present invention also provides a system for implementing the above method, comprising: The data sensing module is used to establish communication connections with photovoltaic inverters, battery management systems (BMS), charging pile group controllers and smart meters of the grid, collect real-time power and battery status data, and preprocess and normalize the collected data. The storage module is used to store the battery aging cost quantification model and the trained deep reinforcement learning neural network parameters; The decision control module is equipped with a processor to run deep reinforcement learning algorithms and calculate the optimal charging and discharging power command based on the input state data. The execution module communicates with the converter (PCS) of the energy storage system to receive power commands, verify and execute them, and feed back the execution results to the data sensing module to form a closed-loop control.

[0018] Furthermore, the present invention also provides an electronic device for implementing the above method, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement the steps of the above-described energy management method for an integrated photovoltaic-storage-charging station that takes into account the battery health status.

[0019] The beneficial effects of this invention are as follows: This invention solves the problem of the inverted life-cycle cost of energy storage systems in high-power supercharging scenarios by deeply integrating mechanistic models and data-driven approaches, coordinating economic objectives and lifespan constraints, and providing dual insurance of intelligent decision-making and functional safety. It achieves a triple paradigm leap from "single electricity price arbitrage" to "dynamic lifespan-revenue trade-off," from "rule-based logic" to "adaptive intelligence," and from "discrete control" to "continuous smoothing." While meeting the supercharging needs of electric vehicles, it effectively suppresses the rapid lifespan degradation of energy storage batteries caused by high-rate charging and discharging, achieving synergistic optimization of economic benefits and equipment lifespan.

[0020] (1) This invention introduces a nonlinear aging cost based on the rate and charge / discharge depth into the reward function, enabling the algorithm to intelligently perceive the charge / discharge state. In non-emergency situations, the agent will automatically avoid high-rate charge / discharge, thereby effectively slowing down the decay rate of SOH and reducing the full life cycle operation and maintenance cost of the energy storage system. This fundamentally solves the problem of imbalance between short-term gains and long-term losses caused by neglecting depreciation costs in traditional algorithms.

[0021] (2) This invention uses a multi-objective weighted reward function to dynamically weight and integrate the grid interaction cost, monetized aging cost and system default penalty cost, so that the system can autonomously weigh the "low storage and high release arbitrage opportunity" and the "high rate of life loss" at multiple time scales, rather than pursuing the maximization of a single electricity price difference, and achieve a dynamic balance between electricity revenue and equipment depreciation.

[0022] (3) The Deep Reinforcement Learning (DDPG) algorithm used has strong adaptive capabilities, can handle the strong randomness of photovoltaic and supercharging loads, and supports continuous power control. Compared with discrete control strategies, it can adjust the system power more smoothly and reduce fluctuations.

[0023] (4) The Deep Deterministic Strategy Gradient (DDPG) algorithm used in this invention breaks through the limitation of the discrete action space, supports the direct output of continuous power commands, and can realize smooth and shock-free adjustment of energy storage power, avoiding the power step shock caused by the discretization of actions in traditional Q-learning and other algorithms. At the same time, by introducing process noise and experience playback mechanism, it can effectively improve the handling of the dual strong randomness of photovoltaic power output and supercharging load.

[0024] (5) Based on intelligent decision-making, this invention integrates physical safety boundary verification, which can intercept and correct dangerous working conditions such as overcharging, over-discharging, power exceeding limits, and rate exceeding limits. It achieves engineering compatibility between "intelligence" and "safety", and significantly improves the robustness and commercial viability of the system under complex working conditions. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the hardware architecture of the integrated photovoltaic, energy storage, and charging station system in an embodiment. Figure 2 This is the main flowchart of the energy management method considering battery health status in the embodiment; Figure 3 This is a logic block diagram of the deep reinforcement learning control strategy in the embodiment; Figure 4 This is a graph showing the nonlinear relationship between battery aging cost and charge / discharge rate in the embodiment. Figure 5 This is a simulation comparison of the energy management strategy and the traditional strategy in terms of battery life degradation in the embodiment. Detailed Implementation

[0026] To make the technical problems and solutions solved by this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Example 1

[0027] To illustrate the invention more concretely, the following description uses a specific scenario of a highway photovoltaic-storage-charging integrated station as an example. In this embodiment, the basic hardware parameters of the station are set as follows: the photovoltaic array installed capacity is 250kW; the energy storage system adopts a rated capacity... It uses lithium iron phosphate batteries, and the bidirectional converter has a maximum allowable charge and discharge power of 400kW; the maximum instantaneous output power of a single supercharging station can reach 360kW.

[0028] like Figure 2 As shown, this embodiment uses the energy management method for the integrated photovoltaic-energy storage-charging station described in this invention to manage the aforementioned integrated photovoltaic-energy storage-charging station on highways, as detailed below: S1: Data Acquisition and Preprocessing In this embodiment, the integrated photovoltaic-storage-charging station energy management system first establishes communication connections with the photovoltaic inverter, the battery management system (BMS), the charging pile, and the smart meter of the power grid through the central controller.

[0029] Specifically, the system uses a fixed time step. A control cycle consists of 5 minutes, during which the system's operational status data is collected in real time. The operational status data includes: Photovoltaic data: Real-time power generation of the current photovoltaic array ; Load-side data: Total power demand of electric vehicles currently connected to charging stations This data can be obtained by summarizing the BMS demand instructions uploaded by the charging pile; Energy storage side data: Current state of charge of the energy storage battery Current health status ; Grid-side data: Current electricity trading price .

[0030] Data normalization processing: Because the dimensions of the above data differ greatly, directly inputting it into the neural network would lead to gradient explosion or difficulty in convergence. Therefore, before proceeding to step S3, the controller performs maximum-minimum normalization on the above state data, mapping it to the interval [0, 1].

[0031] S2: Constructing a battery aging cost model In this embodiment, in order to accurately quantify the SOH loss of the energy storage battery under different operating conditions, a semi-empirical aging model based on electrochemical mechanisms was constructed, which specifically includes the following sub-steps: 1. Extraction of charge-discharge cycle characteristics (based on rainflow counting method) In actual operation, the SOC (State of Charge) curve of a photovoltaic-energy storage-charging system is an irregular, fluctuating curve. To identify effective charge-discharge cycles, this invention uses rainflow counting to process historical or simulated SOC data streams.

[0032] Specifically, the rainflow counting method maps SOC time series data to the trajectory of raindrops flowing down the eaves. By identifying closed hysteresis loops, it extracts the depth of discharge (DOD), mean SOC, and duration of each effective cycle. This step solves the technical problem that traditional methods cannot define "one cycle" under random fluctuation conditions.

[0033] 2. Establish a capacity decay model (based on the Arrhenius model). After extracting the cycling characteristics, an improved Arrhenius semi-empirical model was used to calculate the battery's capacity decay rate. The Arrhenius equation, by incorporating the relationship between temperature and reaction rate, can effectively describe the variation of internal side reactions of the battery under different operating conditions.

[0034] In this embodiment, the battery capacity decay rate Represented as:

[0035] in, This is the forward exponent factor, which is related to the charge / discharge rate; Activation energy; It is the ideal gas constant; This refers to the absolute temperature of the battery. Accumulate and process power for the battery; This is the power-law exponent.

[0036] 3. Converted into the single-step aging cost of reinforcement learning In order to integrate the above-mentioned complex physical model into the real-time control of deep reinforcement learning, the present invention discretizes the above model.

[0037] The system pre-calculates the battery cycle life by using the aforementioned rainflow counting method and Arrhenius model at different charge / discharge rates through offline simulation. With charge / discharge rate The functional relationship is such that, in each decision step of reinforcement learning, the agent does not need to run complex rainflow counting in real time, but directly calls the fitted function to calculate the instantaneous aging cost corresponding to the current action. .like Figure 4 As shown, in this embodiment, the cost of a single battery aging cycle exhibits a non-linear exponential relationship with the charge / discharge rate. Figure 4 It can be seen that when Below 0.5C, the curve is flat, and the aging cost is extremely low; while when When the aging cost exceeds 1C or even reaches 1.5C, the aging cost per action increases exponentially.

[0038] S3: Constructing a Deep Reinforcement Learning Decision Framework In a preferred embodiment of the present invention, to address the problem of neglecting the lifespan reduction caused by high-power overcharging in the prior art, the present invention designs a multi-objective comprehensive reward function. This reward function guides the agent to maximize economic benefits while forcibly introducing a state-of-the-art (SOH) decay penalty term.

[0039] Specifically, at time The instantaneous reward value obtained by the Agent Defined as the negative of the total operating cost (i.e., maximizing the reward equals minimizing the cost), its mathematical expression is as follows:

[0040] in, Indicates the cost of grid interaction. This indicates the cost of battery aging and degradation. This indicates the system's penalty cost for breach of contract. These are the corresponding weighting coefficients. In this embodiment, they are set as follows: , , This gives significant weight to penalties for breach of contract in order to ensure the essential demand for charging.

[0041] The specific calculation methods for each item are explained in detail below: 1. Grid interaction costs

[0042] This is intended to guide the system to engage in peak-valley arbitrage with "low storage and high release" and to prioritize the consumption of photovoltaic power.

[0043]

[0044] in, For a moment Power grid exchange capacity (positive value for electricity purchase, negative value for electricity sale); For a moment Electricity price; To control the time step of the cycle.

[0045] 2. Battery aging and depreciation costs

[0046] Traditional methods often overlook this factor, leading to overly aggressive strategies. This invention introduces a dynamic aging cost calculation based on the charge-discharge rate (C-rate). Since the high current in overcharge scenarios is the main cause of SOH decline, this model employs a non-linear aging penalty mechanism.

[0047]

[0048] in, The total purchase cost or replacement cost (in yuan) of the energy storage battery system; Rated capacity (kWh) of the energy storage battery; For a moment Battery charging and discharging power; To achieve the current charge / discharge rate Below is the maximum number of cycles that a battery can theoretically support.

[0049] To demonstrate the physical characteristic that "high-rate driving damages batteries," this embodiment... The model is fitted using the following power function:

[0050] in, This refers to the current charge / discharge rate or cycle depth. and This represents the battery aging characteristic constant.

[0051] Technical effect description: Using this formula, when the Agent performs high-power fast charging and discharging, It will decrease sharply, leading to A surge. This will train the agent to avoid high-rate operations in non-emergency situations, thereby extending battery life.

[0052] 3. System default penalty costs

[0053] To prevent the agent from refusing to charge the electric vehicle to protect the battery, the following compound penalty function is set.

[0054] in, Penalty for power shortage: When the system provides power Unable to meet the charging needs of electric vehicles When applying a power outage penalty coefficient ; Penalizes overcharge of energy storage batteries to prevent overcharging; This is a penalty for over-discharge of the energy storage battery's state of charge (SOC) to prevent over-discharge.

[0055] S4: Model Training Based on DDPG Algorithm like Figure 3 As shown in the logic block diagram, in this embodiment, the Deep Deterministic Policy Gradient (DDPG) algorithm is used to train the policy model offline, demonstrating the closed-loop interaction process between the "environment (including the optical storage and charging system and the reward calculator)" and the "agent DDPG architecture".

[0056] The DDPG algorithm consists of four neural networks: the current Actor network, the target Actor network, the current Critic network, and the target Critic network.

[0057] The specific training process is as follows: 1. Initialization: Initialize the experience replay pool and randomly initialize the parameters of the Actor and Critic networks.

[0058] 2. Agent Interaction: Agent based on current state Output actions through the Actor network (i.e., energy storage charging and discharging power); To prevent the model from getting trapped in local optima, in action Add random noise to increase randomness; System Execution Action The environment transitions to the next state. And calculate the instant reward according to the reward function described in step S3. .

[0059] like Figure 3 As shown on the right, the quadruple Stored in the experience replay pool.

[0060] 3. Parameter update: Once the amount of data in the experience replay pool reaches a preset threshold, a batch of data is randomly sampled for training.

[0061] Update the Critic network: Minimize the loss function so that the Q-value estimated by the Critic network is as close as possible to the true target Q-value (including the immediate reward and the estimated value at the next time step).

[0062] Update the Actor network: Calculate the policy gradient using the chain rule, adjust the Actor network parameters to maximize the Q-value of the Critic network output.

[0063] 4. Target network soft update: A soft update strategy is adopted to slowly update the parameters of the target network to ensure the stability of the training process.

[0064] 5. Iteration: Repeat the above steps until the model converges, that is, the cumulative reward value tends to stabilize.

[0065] S5: Online Control and Secure Execution After model training is complete, the trained Actor network parameters are deployed to the central controller of the integrated photovoltaic-storage-charging station for real-time energy management. The specific execution logic is as follows: 1. Status Awareness: The controller reads the current status in real time. The data was processed and normalized.

[0066] 2. Strategy reasoning: The normalized state vector is input into the Actor network, and the network directly outputs continuous action values.

[0067] 3. Denormalization: Restores the output action value to the actual power command. .

[0068] 4. Set up security boundary verification: If the current state of charge of the energy storage battery And the generated charge / discharge power command is in the charging direction ( Then, the power command will be forcibly modified. =0; If the current state of charge of the energy storage battery Furthermore, the generated charge / discharge power command is in the discharge direction. Then force correction of power command =0; If the absolute value of the generated charging / discharging power command Exceeding the converter's maximum allowable power Then, the amplitude will be limited to ± .

[0069] 5. Command Issuance: The final power command after verification is sent to the PCS to control the operation of the energy storage battery, thereby achieving comprehensive control to smooth photovoltaic fluctuations, meet supercharging requirements, and extend battery life.

[0070] To more clearly illustrate the difference between the method of this invention and traditional methods, a specific single-step decision-making scenario is given in conjunction with the basic parameters of this embodiment: Assuming that during peak electricity price hours at 2 PM on a certain day, the photovoltaic power generation capacity is... An electric vehicle connected to generate The system has a power gap of 300kW to meet the instantaneous supercharging demand.

[0071] 1. Using traditional control strategies: In order to save on the high cost of purchasing electricity during peak periods, traditional strategies will instruct the battery to discharge at full load to fill the gap, that is, output 300kW (equivalent to 1.5C rate).

[0072] 2. Adopting the strategy of this invention: combining Figure 4 Based on the aging cost assessment, Critic network calculations revealed that the deep depreciation cost of the battery due to 1.5C discharge far outweighs the electricity cost savings. Therefore, after comprehensive consideration by the Actor network, the optimal output instruction is likely to be... (Reducing the discharge rate to 0.5C for gentle operation), the remaining 200kW shortfall is supplemented by the grid. Through this dynamic adaptive adjustment, the system actively avoids extreme high-rate operating conditions.

[0073] Furthermore, to verify the overall effectiveness of the invention throughout its entire lifecycle, a simulation of 2000 consecutive charge-discharge cycles was conducted on the system, and the results are as follows: Figure 5 As shown.

[0074] Figure 5 The dashed line represents the traditional strategy that aims solely at minimizing electricity costs. Under frequent overcharging power surges, it passively undergoes high-rate throughput, causing a rapid decline in State of Charge (SOH). After 2000 cycles, it has fallen to approximately 60%, reaching the critical point for retirement. Figure 5The solid line represents the DDPG strategy of this invention. Because the algorithm "senses" and avoids high-loss behaviors through the reward function in each decision step, its SOH decay curve is significantly slowed down, remaining at around 80% after 2000 cycles. This intuitively and fully demonstrates that this invention effectively extends battery life while ensuring supercharging needs. Example 2

[0075] This embodiment provides a hardware system architecture for implementing the method described in Embodiment 1, which achieves closed-loop management of data acquisition, model storage, intelligent decision-making, and execution control through modular design.

[0076] like Figure 1 As shown, the energy management system described in this embodiment adopts a hierarchical distributed structure, including a central controller and a layer of field devices connected to it. The central controller is connected to the cloud platform via industrial Ethernet to achieve remote monitoring and model updates. The various models in the system are interconnected through a high-speed internal bus (such as a gigabit Ethernet bus) to ensure real-time data transmission.

[0077] The data sensing module, serving as the system's front-end interface, is responsible for establishing multi-protocol communication connections with photovoltaic inverters, battery management systems (BMS), charging pile group controllers, and smart meters in the grid, thereby collecting the following data in real time: Photovoltaic side: Connect to the photovoltaic inverter via RS485 bus to collect real-time power generation data; Load side: Connect to the charging pile group controller via CAN bus or Ethernet (TCP / IP) to obtain the charging power demand of each charging pile and the vehicle BMS status in real time, and summarize to obtain the total power demand. Energy storage side: Connects to the energy storage BMS via CAN 2.0B bus to collect the current state of charge of the energy storage battery. Current health status ; On the grid side: Real-time electricity prices are obtained by connecting to the grid dispatching system via 4G / 5G wireless modules or fiber optic Ethernet. .

[0078] The data perception module has a built-in data processing unit (FPGA) that filters and removes outliers from the collected raw data, performs maximum-minimum normalization processing according to the method described in S1, and transmits the processed state vector to the decision control module via DMA (direct memory access).

[0079] Storage module: used to store the battery aging cost quantification model and the trained deep reinforcement learning neural network parameters; Decision control module: As the core computing unit of the system, it is equipped with a high-performance embedded processor to run deep reinforcement learning algorithms and calculate the optimal charging and discharging power command based on the input state data; The execution module, as the system's output interface, communicates with the converter (PCS) of the energy storage system. It receives power commands, verifies and executes them, and feeds back the execution results to the data sensing module to form a closed-loop control.

[0080] The workflow of the above system is as follows: The data sensing module synchronously collects data from the photovoltaic side, load side, energy storage side, and grid side, processes the data to generate a state vector, and sends it to the decision control module via the internal bus. The decision control module loads the pre-stored battery aging cost model and trained deep reinforcement learning neural network parameters from the storage module, executes forward thrust, and generates actions. (i.e., energy storage charging and discharging power) instructions, and calculate the various costs (grid interaction costs) in the reward function. Battery aging and loss costs System default penalty costs After receiving the energy storage charging and discharging power command, the execution module performs safety checks (overcharge / over-discharge / over-power checks) and generates a power command to send to the PCS. Simultaneously, the execution results (actual power, SOC change) are fed back to the data sensing module, forming a closed loop.

[0081] This embodiment also provides an electronic device, namely the aforementioned central controller, which includes: a processor, a memory, and a computer program stored in the memory and executable on the processor.

[0082] The processor can be a central processing unit, an application-specific integrated circuit, a digital signal processor, or a field-programmable gate array.

[0083] The memory can be volatile or non-volatile.

[0084] When the processor executes the computer program, it implements each step of the energy management method for the integrated photovoltaic-storage-charging station mentioned in Embodiment 1 above.

[0085] The present invention has been described in detail above through specific and preferred embodiments. However, those skilled in the art should understand that the present invention is not limited to the embodiments described above. Any modifications, equivalent substitutions, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An energy management method for an integrated photovoltaic, energy storage, and charging station, characterized in that, Includes the following steps: S1. Data Acquisition and Preprocessing: Real-time acquisition of the operating status data of the photovoltaic-storage-charging integrated charging station system. The operating status data includes at least the real-time photovoltaic power generation, electric vehicle charging demand, current state of charge of the energy storage battery, current health status of the energy storage battery, and grid transaction price. The operating status data is then normalized. S2. Construct a battery aging cost quantification model: Based on the rainflow counting method and the Arrhenius semi-empirical model, establish a mapping relationship model between battery life loss and charge / discharge behavior. This model is a battery aging cost quantification model suitable for deep reinforcement learning. Input the charge / discharge power or rate into the model to calculate the instantaneous battery aging cost corresponding to a single action. S3. Constructing a deep reinforcement learning decision-making framework: Define a state space that contains the normalized running state data from step S1; Define the action space as the charging and discharging power commands for the energy storage battery within a continuous interval; Define a reward function, which is configured to maximize the overall system benefit. Its components include at least grid interaction cost, battery aging loss cost obtained based on the single-step aging cost calculation model, and system default penalty cost. S4. Model Training: The Deep Deterministic Policy Gradient (DDPG) algorithm is used to train the policy network offline. The policy network outputs continuous charging and discharging power actions based on the input state vector. The cumulative reward is maximized through an experience replay mechanism and network parameter update strategy. S5. Online Control and Safe Execution: The trained policy network is deployed to the central controller of the charging station. Based on the real-time collected operating status data, charging and discharging power commands are generated and safety boundary verification is performed. Finally, the verified power commands are sent to the energy storage converter for execution.

2. The energy management method for an integrated photovoltaic-storage-charging station according to claim 1, characterized in that, In step S2, the process of constructing the battery aging cost quantification model includes: Rainflow counting is used to process historical or simulated battery state-of-charge time series data to identify closed hysteresis loops and extract the discharge depth, cycle mean, and duration characteristics of effective charge-discharge cycles. A semi-empirical Arrhenius model was used to establish the functional relationship between battery capacity decay rate and charge / discharge rate, depth of discharge, and temperature. Based on the above functional relationship, offline simulations were performed at different charge and discharge rates to obtain a nonlinear function of battery cycle life and charge and discharge rate, which is the battery aging cost quantification model.

3. The energy management method for an integrated photovoltaic-storage-charging station according to claim 2, characterized in that, The battery capacity decay rate The expression is: ； in, This is the forward exponent factor, which is related to the charge / discharge rate; Activation energy; It is the ideal gas constant; This refers to the absolute temperature of the battery. Accumulate and process power for the battery; This is the power-law exponent.

4. The energy management method for an integrated photovoltaic-storage-charging station according to claim 1, characterized in that, In step S3, the reward function The expression is as follows: ； in, Indicates the cost of grid interaction; The battery aging loss cost is represented by the battery aging cost quantification model obtained through step S2. This represents the system's default penalty cost, which includes at least the power shortage penalty for failing to meet the charging needs of electric vehicles and the penalty for exceeding the SOC limit of the energy storage battery. These are the corresponding weighting coefficients.

5. The energy management method for an integrated photovoltaic-storage-charging station according to claim 1, characterized in that, In step S4, the specific process of offline training using the Deep Deterministic Policy Gradient (DDPG) algorithm is as follows: (1) Initialize the experience replay pool, and initialize the current network of the Actor, the target network of the Actor, the current network of the Critic, and the target network of the Critic; (2) The Agent outputs the energy storage charging and discharging power action through the Actor network according to the current state, and adds random noise to the action to increase randomness; (3) After the system performs the action, the environment is transferred to the next state, and the immediate reward is calculated according to the reward function described in step S3. The quadruple consisting of state, action, reward and next state is stored in the experience replay pool. (4) When the amount of data in the experience replay pool reaches the preset threshold, a small batch of data is randomly sampled, the parameters of the Critic network are updated to minimize the temporal difference error, and the parameters of the Actor network are updated to maximize the Q value of the Critic network output. (5) Use a soft update strategy to slowly update the target network parameters until the cumulative reward value tends to stabilize.

6. The energy management method for an integrated photovoltaic-storage-charging station according to claim 1, characterized in that, In step S5, the security boundary verification specifically includes: If the current state of charge of the energy storage battery And the generated charge / discharge power command is in the charging direction ( Then, the power command will be forcibly modified. =0; If the current state of charge of the energy storage battery Furthermore, the generated charge / discharge power command is in the discharge direction. Then force correction of power command =0; If the absolute value of the generated charging / discharging power command Exceeding the converter's maximum allowable power Then, the amplitude will be limited to... .

7. The energy management method for an integrated photovoltaic-storage-charging station according to claim 1, characterized in that, The electric vehicle charging power demand includes the high-power random load of the supercharging piles; during the DDPG algorithm training process, random noise is superimposed on the supercharging pile load power to simulate the uncertainty of the supercharging load and improve the algorithm's adaptability to high-rate charging and discharging conditions.

8. An integrated photovoltaic-storage-charging station energy management system, used to implement the method described in any one of claims 1-7, characterized in that, include: The data sensing module is used to establish communication connections with photovoltaic inverters, energy storage battery management systems, charging pile group controllers and smart meters of the power grid, collect real-time power and battery status data, and preprocess and normalize the collected data. The storage module is used to store the battery aging cost quantification model and the trained deep reinforcement learning neural network parameters; The decision control module is equipped with a processor to run deep reinforcement learning algorithms and calculate the optimal charging and discharging power command based on the input state data. The execution module communicates with the converter of the energy storage system to receive power commands, verify and execute them, and feed the execution results back to the data sensing module to form a closed-loop control.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the energy management method for the integrated photovoltaic-storage-charging station according to any one of claims 1-7.

Images