Methods and Systems for Collaborative Scheduling and Control of Virtual Power Plant Energy Storage Systems

Abstract

This invention provides a collaborative scheduling and control method and system for a virtual power plant energy storage system, relating to the field of power management technology. The method includes: acquiring grid dispatch data to calculate charging and discharging benefits; collecting real-time battery parameters; applying deep reinforcement learning to analyze performance degradation patterns and determine initial operating parameters; predicting system state based on rolling time domain; combining power allocation strategies with mixed-integer linear programming and model predictive control to calculate and correct the optimal scheduling scheme in real time; executing operations according to time-segmented scheduling instructions and monitoring in real time; and activating emergency strategies when necessary. This improves the economic efficiency and safety of the energy storage system and extends battery life.

Description

Technical Field

[0001] This invention relates to the field of power management technology, and in particular to a method and system for collaborative scheduling and control of a virtual power plant energy storage system. Background Technology

[0002] With the transformation of energy structure and the large-scale integration of renewable energy, the power system faces challenges in many aspects, such as supply and demand balance, grid stability and energy utilization efficiency. As a new energy management model, virtual power plants integrate and dispatch dispersed energy resources through information and communication technologies, which can effectively improve system flexibility and reliability. As a core component of virtual power plants, energy storage systems play an important role in smoothing fluctuations, peak shaving and valley filling, and frequency regulation.

[0003] Traditional energy storage system scheduling mainly relies on simplified models for optimization decisions. However, it still ignores the performance and degradation characteristics of individual battery cells, resulting in uneven utilization of battery components, difficulty in coping with grid fluctuations and load changes, inability to achieve globally optimal control, difficulty in dynamically adjusting control strategies according to actual state changes during operation, insufficient system safety, and inability to respond in a timely manner, especially under extreme conditions. Summary of the Invention

[0004] This invention provides a method and system for collaborative scheduling and control of a virtual power plant energy storage system, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides a method for collaborative scheduling and control of a virtual power plant energy storage system, comprising:

[0006] Real-time operating parameters of individual battery clusters in the energy storage system are collected, and the real-time operating parameters are analyzed based on deep reinforcement learning methods to obtain the performance degradation law of individual battery clusters. The initial operating parameters of each individual battery cluster are determined based on the performance degradation law.

[0007] Based on the rolling time domain, the system operating status of multiple future time periods is predicted to obtain the system status prediction value. The power allocation strategy is dynamically generated according to the system status prediction value and the initial operating parameters. The optimal scheduling scheme for each time period is calculated by combining the power allocation strategy with mixed integer linear programming and model predictive control methods. The optimal scheduling scheme is corrected in real time by comprehensively considering grid fluctuations, load changes and charging and discharging benefits to obtain the time-sharing scheduling instructions of the energy storage system.

[0008] The system executes charging and discharging operations according to the time-segmented scheduling instructions and monitors the operating parameters of the energy storage system in real time. When the operating parameters exceed the preset safety threshold, the charging and discharging power is adjusted according to the preset emergency control strategy.

[0009] In one alternative implementation,

[0010] Obtain grid dispatch data within a preset time period and calculate the charging and discharging benefits of the energy storage system under system constraints, including:

[0011] Obtain grid load demand information and electricity price information within a preset time period;

[0012] A time-series analysis of the power grid load demand information and electricity price information is performed to obtain the charging and discharging period division results, which include high load and high electricity price periods and low load and low electricity price periods;

[0013] Based on the results of the charging and discharging time period division, and combined with the power and capacity constraints of the energy storage system, charging operations are performed during low-load and low-electricity-price periods, and discharging operations are performed during high-load and high-electricity-price periods, to calculate the charging and discharging revenue of the energy storage system.

[0014] In one alternative implementation,

[0015] Real-time operating parameters of individual battery cells in the energy storage system are collected. These parameters are then analyzed using deep reinforcement learning to determine the performance degradation patterns of the individual battery cells. Based on these performance degradation patterns, the initial operating parameters of each individual battery cell are determined, including:

[0016] Real-time operating parameters of individual battery cells are collected, and outliers are removed using a three-standard-deviation criterion. The outliers are then standardized by maximum and minimum values. Wavelet transform is used to extract time-domain features from the standardized real-time operating parameters to obtain the characteristic parameters of individual battery cells. The real-time operating parameters include terminal voltage parameters, open-circuit voltage parameters, battery state of charge, cycle number, surface temperature parameters, and internal temperature distribution parameters.

[0017] The feature parameters are analyzed based on a deep reinforcement learning algorithm to obtain the capacity decay law and internal resistance growth law of individual battery clusters, and the capacity retention rate and internal resistance change rate of individual battery clusters are calculated. The health status of individual battery clusters is determined based on the capacity retention rate and internal resistance change rate.

[0018] The charge / discharge depth limit is determined based on the health status and the real-time operating parameters. The power of each battery cluster cell is allocated based on the charge / discharge depth limit to obtain the initial operating parameters of each battery cluster cell.

[0019] In one alternative implementation,

[0020] The feature parameters are analyzed using a deep reinforcement learning algorithm to obtain the capacity decay law and internal resistance growth law of individual battery cluster cells, and the capacity retention rate and internal resistance change rate of individual battery cluster cells are calculated, including:

[0021] A temperature field gradient conduction network is established based on the aforementioned characteristic parameters. The spatial temperature gradient and time-varying temperature change rate of the surface temperature parameter and the internal temperature distribution parameter among the characteristic parameters are calculated. The temperature distribution characteristics of individual cells in the battery cluster are determined based on the spatial temperature gradient and the time-varying temperature change rate.

[0022] A thermal conduction matrix is ​​established based on the physical positional relationship of the individual cells in the battery cluster. The thermal resistance between cells is calculated and a temperature field conduction equation is established to obtain the thermal flow conduction characteristics of the individual cells in the battery cluster. The thermal stress is calculated based on the thermal flow conduction characteristics and the stress distribution characteristics of the electrode material are determined. The stress distribution characteristics are coupled with the temperature distribution characteristics to obtain the temperature-stress coupling characteristics.

[0023] The spatial temperature gradient, time-temperature change rate, and temperature stress coupling features are fused with the feature parameters to obtain a fused feature vector. Based on the deep reinforcement learning algorithm, the fused feature vector is analyzed to establish a temperature field capacity decay coupling model and a temperature field internal resistance growth coupling model, thereby obtaining the capacity decay law and internal resistance growth law of the battery cluster cells.

[0024] The initial capacity retention rate and initial internal resistance change rate of the individual cells in the battery cluster are calculated based on the capacity decay law and internal resistance growth law. The initial capacity retention rate and initial internal resistance change rate are then corrected by the spatial temperature gradient and temperature stress coupling characteristics to obtain the capacity retention rate and internal resistance change rate of the individual cells in the battery cluster.

[0025] In one alternative implementation,

[0026] Based on rolling time domain prediction of system operating states for multiple future time periods, system state prediction values ​​are obtained. A power allocation strategy is then dynamically generated based on these system state prediction values ​​and the initial operating parameters, including:

[0027] A rolling prediction interval and a rolling update cycle are set. A deep learning algorithm is used to predict the load power, grid power, energy storage power and state of charge within the rolling prediction interval. The predicted values ​​of load power, grid power, energy storage power and state of charge for each time period are obtained and combined to obtain the predicted value of system state.

[0028] Based on the system state prediction value and the initial operating parameters, calculate the maximum allowable fluctuation range of grid power and energy storage power, the maximum allowable charging and discharging range of energy storage power, and the maximum allowable exchange range of grid power, and generate an initial power allocation scheme.

[0029] The initial power allocation scheme is applied to the grid power forecast and energy storage power forecast for each time period to ensure that the sum of the grid power and energy storage power for each time period is equal to the load power forecast for the corresponding time period. The energy storage power is within the maximum allowable charge and discharge range, and the grid power is within the maximum allowable exchange range, thus obtaining the power allocation strategy.

[0030] In one alternative implementation,

[0031] By employing mixed-integer linear programming and model predictive control methods, combined with the power allocation strategy, the optimal scheduling scheme for each time period is calculated. Taking into account grid fluctuations, load changes, and charging / discharging benefits, the optimal scheduling scheme is then corrected in real time, resulting in time-segmented scheduling instructions for the energy storage system, including:

[0032] Extract power time-series data to calculate power fluctuation sequence, extract load power time-series data to calculate load change sequence, construct the current power grid power and load power as the root node state vector, superimpose the power grid fluctuation sequence and load change sequence onto the root node state vector at preset time intervals to obtain child node state vectors at different times, construct the root node state vector and child node state vectors into a scene tree structure, and calculate the occurrence probability of each node in the scene tree structure and the transition probability between adjacent nodes;

[0033] Based on the scenario tree structure, establish a continuous variable vector containing energy storage power variables and state of charge variables, establish an integer variable vector containing charging state indicator variables and discharging state indicator variables, establish a power balance constraint condition that the sum of grid power and energy storage power equals the load power, and establish energy storage system operation constraints for the range of energy storage power and state of charge values.

[0034] The continuous variable vector and integer variable vector are linearized. Under the constraints of the power balance constraint and the energy storage system operation constraint, the energy storage power and charge / discharge state corresponding to the minimum cost are calculated to obtain the initial optimal solution. A prediction time window and a control time window are set. Within the prediction time window, the predicted values ​​of the grid power and load power for future periods are calculated based on the current grid power and load power. Within the control time window, the energy storage power and charge / discharge state for each period are calculated. The power allocation strategy is used as a constraint to optimize and iteratively calculate the energy storage power and charge / discharge state to obtain the optimal scheduling scheme for each period.

[0035] Calculate the deviation between the current grid power and load power and the grid power and load power in the sub-node state vector. Based on the deviation, calculate the scenario weight coefficient and power correction amount. Combine the grid fluctuation sequence, load change sequence and the charging and discharging benefits, and superimpose the power correction amount with the energy storage power in the optimal scheduling scheme to obtain the time-sharing scheduling instruction of the energy storage system.

[0036] In one alternative implementation,

[0037] According to the time-segmented scheduling instructions, charging and discharging operations are performed, and the operating parameters of the energy storage system are monitored in real time. When the operating parameters exceed a preset safety threshold, the charging and discharging power is adjusted according to a preset emergency control strategy, including:

[0038] Receive time-segmented scheduling instructions, and control the energy storage system to perform charging and discharging operations according to the energy storage power and charging / discharging status in the time-segmented scheduling instructions;

[0039] Monitor the real-time operating parameters of the energy storage system and compare them with the corresponding preset safety thresholds;

[0040] When any parameter exceeds the corresponding preset safety threshold, the charging power or discharging power is reduced according to the power adjustment rules in the preset emergency control strategy until the operating parameter returns to the preset safety threshold range.

[0041] A second aspect of the present invention provides a collaborative scheduling and control system for a virtual power plant energy storage system, comprising:

[0042] The first unit is used to acquire grid dispatch data within a preset time period and calculate the charging and discharging benefits of the energy storage system under the condition of meeting system constraints.

[0043] The second unit is used to collect real-time operating parameters of individual battery clusters in the energy storage system, analyze the real-time operating parameters based on deep reinforcement learning methods, obtain the performance degradation law of individual battery clusters, and determine the initial operating parameters of each individual battery cluster based on the performance degradation law.

[0044] The third unit is used to predict the system operating status for multiple future time periods based on the rolling time domain to obtain the system status prediction value. Based on the system status prediction value and the initial operating parameters, a power allocation strategy is dynamically generated. By using mixed integer linear programming and model predictive control methods, combined with the power allocation strategy, the optimal scheduling scheme for each time period is calculated. Taking into account grid fluctuations, load changes and charging and discharging benefits, the optimal scheduling scheme is corrected in real time to obtain the time-sharing scheduling instructions for the energy storage system.

[0045] The fourth unit is used to perform charging and discharging operations according to the time-sharing scheduling instructions and monitor the operating parameters of the energy storage system in real time. When the operating parameters exceed the preset safety threshold, the charging and discharging power is adjusted according to the preset emergency control strategy.

[0046] A third aspect of the present invention provides an electronic device, comprising:

[0047] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0048] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0049] In this invention, the performance degradation law of individual battery cells in the energy storage system is analyzed by deep reinforcement learning. The initial operating parameters are determined based on the degradation law, which effectively extends the battery life and improves the overall reliability of the system. The power allocation strategy is dynamically generated and the optimal scheduling scheme is calculated. Taking into account grid fluctuations, load changes and charging and discharging benefits, the efficient scheduling and optimal operation of the energy storage system are achieved, which significantly improves economic benefits. When the operating parameters exceed the safety threshold, the charging and discharging power can be adjusted according to the preset emergency control strategy to ensure the safe operation of the energy storage system, enhance the system's ability to respond to emergencies, and improve the stability and reliability of the entire virtual power plant. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the collaborative scheduling and control method for a virtual power plant energy storage system according to an embodiment of the present invention.

[0051] Figure 2 This is a thermal stress distribution diagram of the battery electrode material in the collaborative scheduling and control method of the virtual power plant energy storage system according to an embodiment of the present invention.

[0052] Figure 3 This is a flowchart illustrating the power allocation strategy generation process of the collaborative scheduling and control method for a virtual power plant energy storage system according to an embodiment of the present invention.

[0053] Figure 4 This is a comparison chart of the power grid power balance performance of the collaborative scheduling and control method of the virtual power plant energy storage system in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0056] Figure 1 This is a flowchart illustrating the collaborative scheduling and control method for a virtual power plant energy storage system according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0057] Real-time operating parameters of individual battery clusters in the energy storage system are collected, and the real-time operating parameters are analyzed based on deep reinforcement learning methods to obtain the performance degradation law of individual battery clusters. The initial operating parameters of each individual battery cluster are determined based on the performance degradation law.

[0058] Based on the rolling time domain, the system operating status of multiple future time periods is predicted to obtain the system status prediction value. The power allocation strategy is dynamically generated according to the system status prediction value and the initial operating parameters. The optimal scheduling scheme for each time period is calculated by combining the power allocation strategy with mixed integer linear programming and model predictive control methods. The optimal scheduling scheme is corrected in real time by comprehensively considering grid fluctuations, load changes and charging and discharging benefits to obtain the time-sharing scheduling instructions of the energy storage system.

[0059] The system executes charging and discharging operations according to the time-segmented scheduling instructions and monitors the operating parameters of the energy storage system in real time. When the operating parameters exceed the preset safety threshold, the charging and discharging power is adjusted according to the preset emergency control strategy.

[0060] In one alternative implementation,

[0061] Obtain grid dispatch data within a preset time period and calculate the charging and discharging benefits of the energy storage system under system constraints, including:

[0062] Obtain grid load demand information and electricity price information within a preset time period;

[0063] A time-series analysis of the power grid load demand information and electricity price information is performed to obtain the charging and discharging period division results, which include high load and high electricity price periods and low load and low electricity price periods;

[0064] Based on the results of the charging and discharging time period division, and combined with the power and capacity constraints of the energy storage system, charging operations are performed during low-load and low-electricity-price periods, and discharging operations are performed during high-load and high-electricity-price periods, to calculate the charging and discharging revenue of the energy storage system.

[0065] This system retrieves grid load demand and electricity price information for a preset time period. In practical applications, data can be obtained from power trading centers, grid companies, or energy management systems. For example, it can retrieve the grid load curve and time-of-use electricity price data for a specific region over the next 24 hours. Load demand information is typically recorded hourly or every 15 minutes, showing the grid's electricity load at different times, in kilowatts or megawatts. Electricity price information records the electricity price at different times, in yuan per kilowatt-hour. A data connection is established with the grid dispatch system via an API interface to automatically retrieve updated load and electricity price data periodically and store it in a local database for subsequent analysis.

[0066] After acquiring the data, time-series analysis is performed on the power grid load demand and electricity price information. This includes three steps: data preprocessing, time period feature extraction, and time period segmentation. Missing values ​​are handled, and outliers are detected and corrected in the raw data to ensure data quality. For example, abnormal fluctuations in the load curve are smoothed using a moving average method; missing electricity price data is filled in using the average value of historical data for the same time period.

[0067] In the time period feature extraction stage, a comprehensive index for each time period is calculated. This index takes into account both load level and electricity price level. The load value and electricity price of each time period are divided by their average value to obtain a normalized value. The weighted sum is then calculated as the comprehensive index for that time period. For example, the load data (megawatts) for a certain 24-hour period is: [80, 70, 65, 60, 65, 75, 90, 110, 120, 125, 130, 135, 140, 135, 130, 135, 140, 145, 140, 130, 120, 110, 100, 90], and the corresponding electricity price data (yuan / kWh) is: [0.3, 0.3, 0.3, 0.3, 0.3, 0.4, 0.5, 0.8, 0.8, 0.8, 1.0, 1.0, 1.0, 0.8, 0.8, 0.8, 1.2, 1.2, 1.2, 0.8, 0.8, 0.5, 0.4, 0.3]. After normalization and comprehensive calculation, the comprehensive index value for each hour is generated.

[0068] During the time period segmentation phase, a threshold is set based on the comprehensive index value. Time periods above the high threshold are classified as high load and high electricity price periods (discharging periods), and time periods below the low threshold are classified as low load and low electricity price periods (charging periods). For example, 1:00 AM to 5:00 AM is classified as a low load and low electricity price period (charging period), and 10:00 AM to 2:00 PM and 5:00 PM to 7:00 PM are classified as high load and high electricity price periods (discharging periods).

[0069] After the time periods are divided, based on the results of the charging and discharging time period division, and combined with the power and capacity constraints of the energy storage system, charging operations are performed during low-load, low-electricity-price periods, and discharging operations are performed during high-load, high-electricity-price periods, thereby calculating the charging and discharging revenue of the energy storage system. Power constraints refer to the maximum charging and discharging power limit of the energy storage system, while capacity constraints refer to the available capacity range and depth of charge / discharge limits of the energy storage system.

[0070] Initialize the initial state of charge of the energy storage system, for example, set it to 30% capacity. For each time period, determine whether to charge or discharge based on the time period division. During the charging period, calculate the rechargeable amount, which depends on the remaining capacity and maximum charging power of the energy storage system; during the discharging period, calculate the dischargeable amount, which depends on the current charge and maximum discharging power of the energy storage system, taking into account the charging and discharging efficiency of the energy storage system.

[0071] For example, suppose the energy storage system has a capacity of 1000 kWh, a maximum charging / discharging power of 200 kW, and a charging / discharging efficiency of 90%. At 2 AM (a low-load, low-electricity-price period, with an electricity price of 0.3 yuan / kWh), a charging decision is made. The current stored energy is 350 kWh, so the rechargeable amount is min(200 kW, (1000-350) / 0.9) = 200 kW. After one hour of charging, the stored energy increases to 350 + 200 × 0.9 = 530 kWh, and the charging cost is 200 × 0.3 = 60 yuan. At 6 PM (a high-load, high-electricity-price period, with an electricity price of 1.2 yuan / kWh), a discharging decision is made. The current energy storage capacity is 830 kWh, and the discharge capacity is min(200 kW, 830 × 0.9) = 200 kW. After discharging for one hour, the energy storage capacity decreases to 830 - 200 / 0.9 = 608 kWh, and the discharge revenue is 200 × 1.2 = 240 yuan.

[0072] By accumulating the revenue and cost of all charging and discharging operations within a preset time period, the net revenue of the energy storage system during that time period can be calculated. If the energy storage system charges a total of 600 kWh (cost 180 yuan) and discharges 550 kWh (revenue 605 yuan) in a day, the daily net revenue is 605-180=425 yuan.

[0073] In this embodiment, precise time-series analysis can capture electricity price fluctuation characteristics and maximize the charge-discharge price difference. The decision-making mechanism based on time-series analysis enables the system to adapt to the grid load and electricity price change characteristics in different regions and seasons, exhibiting strong versatility and adaptability. Through the charge-discharge operation of the energy storage system, grid load fluctuations can be effectively smoothed, grid peak pressure can be reduced, grid stability and reliability can be improved, and energy utilization efficiency can be improved, providing strong support for the stable operation of the entire power system and the efficient utilization of renewable energy.

[0074] In one alternative implementation,

[0075] Real-time operating parameters of individual battery cells in the energy storage system are collected. These parameters are then analyzed using deep reinforcement learning to determine the performance degradation patterns of the individual battery cells. Based on these performance degradation patterns, the initial operating parameters of each individual battery cell are determined, including:

[0076] Real-time operating parameters of individual battery cells are collected, and outliers are removed using a three-standard-deviation criterion. The outliers are then standardized by maximum and minimum values. Wavelet transform is used to extract time-domain features from the standardized real-time operating parameters to obtain the characteristic parameters of individual battery cells. The real-time operating parameters include terminal voltage parameters, open-circuit voltage parameters, battery state of charge, cycle number, surface temperature parameters, and internal temperature distribution parameters.

[0077] The feature parameters are analyzed based on a deep reinforcement learning algorithm to obtain the capacity decay law and internal resistance growth law of individual battery clusters, and the capacity retention rate and internal resistance change rate of individual battery clusters are calculated. The health status of individual battery clusters is determined based on the capacity retention rate and internal resistance change rate.

[0078] The charge / discharge depth limit is determined based on the health status and the real-time operating parameters. The power of each battery cluster cell is allocated based on the charge / discharge depth limit to obtain the initial operating parameters of each battery cluster cell.

[0079] During the operation of the energy storage system, the deployed battery management system collects real-time operating parameters of individual battery cells at a frequency of once per second. The collected parameters include terminal voltage, open-circuit voltage, state of charge (SOC), cycle count, surface temperature, and internal temperature distribution. Terminal voltage is measured by voltage acquisition modules connected to the positive and negative terminals of the battery, with a measurement range of 0-5V and an accuracy of ±0.01V. Open-circuit voltage is measured by placing the battery in a static state for at least 30 minutes. SOC is obtained using a coulomb counting method combined with an open-circuit voltage correction method, with an accuracy controlled within ±2%. Cycle count is accumulated and recorded by a counter in the battery management system. Surface temperature is collected using thermistors attached to the battery surface, with acquisition points set at the top, middle, and bottom of the battery, and an accuracy of ±0.5℃. Internal temperature distribution parameters are obtained through a miniature temperature sensor array embedded inside the battery, with a total of 8 temperature measurement points, forming data on the internal temperature field distribution of the battery.

[0080] Outlier removal is performed on the collected real-time operating parameters using a three-standard-deviation criterion. Taking terminal voltage as an example, for 1000 continuously collected data sets, the average value is calculated to be 3.65V, and the standard deviation is 0.08V. Data values ​​outside the range [3.41V, 3.89V] are considered outliers and removed. The same outlier removal method is used for open-circuit voltage, battery state of charge, surface temperature parameters, and internal temperature distribution parameters. For example, if the average battery surface temperature data is 28.5℃, and the standard deviation is 1.2℃, then data outside the range [24.9℃, 32.1℃] are identified as outliers and removed.

[0081] After removing outliers, the data is standardized using maximum and minimum values. Taking the battery state of charge (SCC) as an example, the original data ranges from 10% to 90%, and standardization converts it into a value within the range of 0 to 1. The standardized value is equal to the original value minus the minimum value, divided by the difference between the maximum and minimum values. For example, if the original battery SCC is 65%, the standardized value is (65%-10%) / (90%-10%) = 0.6875.

[0082] After standardization, wavelet transform was used to extract time-domain features from the data. The db4 wavelet basis function was selected, and a five-level wavelet decomposition was performed on the standardized voltage, temperature, and other parameters to extract energy, entropy, and statistical features. The mean, variance, kurtosis, and skewness of the battery terminal voltage parameters were extracted, and the energy proportion of the wavelet coefficients at different scales was calculated. For example, after wavelet transform, the mean feature of the terminal voltage curve of a single battery cell during charging was extracted to be 0.72, the variance feature to be 0.15, and the energy proportions of the first to fifth level wavelet decompositions were 0.08, 0.12, 0.25, 0.35, and 0.20, respectively. These features together constitute the feature parameter set of the individual battery cells in the battery cluster.

[0083] A deep reinforcement learning network architecture is constructed based on the analysis of feature parameters using deep reinforcement learning algorithms. The network consists of two parts: a feature extraction network and a policy network. The feature extraction network employs a four-layer convolutional neural network structure with kernel sizes of 5×5, 3×3, 3×3, and 3×3, and the number of kernels being 32, 64, 128, and 256, respectively. Each convolutional layer is followed by a max-pooling layer and a batch normalization layer. The policy network uses a three-layer fully connected network with 512, 256, and 128 neurons, and the ReLU activation function is used. The network input consists of battery feature parameters, and the output consists of predicted values ​​for the battery capacity decay rate and internal resistance growth rate.

[0084] The network training employed an experience replay mechanism, with an experience pool size of 10,000 and a batch size of 64. During training, the reward function was designed as the negative of the mean squared error between predicted and actual measurements, aiming to maximize the cumulative reward. The initial learning rate was set to 0.001, using the Adam optimizer and a learning rate decay strategy, decreasing to 0.9 times the original value every 50 epochs. The initial exploration rate ε was set to 0.9, linearly decreasing to 0.1 as training progressed. After 5000 training epochs, the model achieved a prediction accuracy of 92.3% on the validation set.

[0085] By using a trained deep reinforcement learning model, the capacity decay and internal resistance growth patterns of individual cells in a battery cluster are analyzed. Based on the prediction results, the battery capacity retention rate, i.e., the ratio of the current capacity to the initial capacity, is calculated. For example, if the initial capacity of a single cell in a battery cluster is 100 Ah, after 200 cycles, the capacity decays to 93 Ah, resulting in a capacity retention rate of 93%. The internal resistance change rate, i.e., the ratio of the current internal resistance to the initial internal resistance, is then calculated. If the initial internal resistance is 5 mΩ and the current internal resistance is 5.8 mΩ, the internal resistance change rate is 116%.

[0086] The health status of individual cells in a battery cluster is determined based on capacity retention rate and internal resistance change rate. The health status assessment criteria are as follows: when the capacity retention rate is greater than 90% and the internal resistance change rate is less than 120%, the health status is "good"; when the capacity retention rate is between 80% and 90% or the internal resistance change rate is between 120% and 150%, the health status is "fair"; when the capacity retention rate is less than 80% or the internal resistance change rate is greater than 150%, the health status is "poor".

[0087] Based on the battery's health status and real-time operating parameters, the depth of charge and discharge limits are determined. For batteries with a "good" health status, the depth of charge and discharge limit is set at 10%-90%; for batteries with a "fair" health status, the depth of charge and discharge limit is adjusted to 20%-80%; and for batteries with a "poor" health status, the depth of charge and discharge limit is further narrowed to 30%-70%.

[0088] Power allocation among individual cells in a battery cluster is based on charge / discharge depth limits. This allocation employs a weighted distribution principle based on the cell's health status, with cells in better health receiving a higher percentage of power. For example, in a system consisting of three battery clusters with health statuses of "good," "moderate," and "poor," the corresponding power allocation ratios are 50%, 30%, and 20%, respectively. If the total system power is 100kW, then each battery cluster will handle 50kW, 30kW, and 20kW of power, respectively. The initial operating parameters for each individual cell in this system are calculated based on operating voltage, operating current, and operating power, ensuring optimal overall system performance while extending battery life.

[0089] In this embodiment, outlier removal is performed using the three-standard-deviation criterion, combined with maximum and minimum value standardization, which effectively improves the data quality of battery operating parameters and reduces the interference of outlier data on subsequent analysis. Wavelet transform is used to extract time-domain features, which can comprehensively capture the dynamic changes of battery operating parameters. Compared with traditional time-domain analysis methods, it can more accurately reflect the changing patterns of battery operating status. Based on deep reinforcement learning algorithms, capacity decay and internal resistance growth patterns are analyzed, achieving adaptive assessment of battery health status. This avoids the limitations of traditional fixed threshold methods and improves assessment accuracy.

[0090] In one alternative implementation,

[0091] The feature parameters are analyzed using a deep reinforcement learning algorithm to obtain the capacity decay law and internal resistance growth law of individual battery cluster cells, and the capacity retention rate and internal resistance change rate of individual battery cluster cells are calculated, including:

[0092] A temperature field gradient conduction network is established based on the aforementioned characteristic parameters. The spatial temperature gradient and time-varying temperature change rate of the surface temperature parameter and the internal temperature distribution parameter among the characteristic parameters are calculated. The temperature distribution characteristics of individual cells in the battery cluster are determined based on the spatial temperature gradient and the time-varying temperature change rate.

[0093] A thermal conduction matrix is ​​established based on the physical positional relationship of the individual cells in the battery cluster. The thermal resistance between cells is calculated and a temperature field conduction equation is established to obtain the thermal flow conduction characteristics of the individual cells in the battery cluster. The thermal stress is calculated based on the thermal flow conduction characteristics and the stress distribution characteristics of the electrode material are determined. The stress distribution characteristics are coupled with the temperature distribution characteristics to obtain the temperature-stress coupling characteristics.

[0094] The spatial temperature gradient, time-temperature change rate, and temperature stress coupling features are fused with the feature parameters to obtain a fused feature vector. Based on the deep reinforcement learning algorithm, the fused feature vector is analyzed to establish a temperature field capacity decay coupling model and a temperature field internal resistance growth coupling model, thereby obtaining the capacity decay law and internal resistance growth law of the battery cluster cells.

[0095] The initial capacity retention rate and initial internal resistance change rate of the individual cells in the battery cluster are calculated based on the capacity decay law and internal resistance growth law. The initial capacity retention rate and initial internal resistance change rate are then corrected by the spatial temperature gradient and temperature stress coupling characteristics to obtain the capacity retention rate and internal resistance change rate of the individual cells in the battery cluster.

[0096] The battery cluster is divided into n×m×l grid cells, each representing a tiny region in space. Each grid cell is assigned a surface temperature value extracted from feature parameters, and a three-dimensional convolutional neural network (CNN) is used to model the temperature field. The CNN consists of five convolutional layers, each using 32 3×3×3 convolutional kernels, with the ReLU activation function. This network calculates the spatial temperature gradient by measuring the temperature difference between adjacent grid cells, with a specific value ranging from 0.5°C / cm to 3.5°C / cm. The temporal temperature change rate is calculated using temperature data at continuous time points, typically ranging from 0.02°C / s to 0.5°C / s under normal operating conditions. For example, in a set of test data, the spatial temperature gradient in the central region of the battery cluster reached 2.8°C / cm, while the edge region had a gradient of 1.2°C / cm, and the maximum temporal temperature change rate during charging and discharging was 0.35°C / s.

[0097] A thermal conductivity matrix is ​​established based on the physical positional relationships of individual cells in the battery cluster. Each cell in the cluster is considered a node, and the physical connections between nodes are considered edges, constructing an undirected weighted graph. The weight of each edge represents the thermal conductivity coefficient between two cells, which is determined by material properties and contact area. For closely packed lithium battery packs, the typical thermal conductivity coefficient between adjacent cells ranges from 0.8 W / (m·K) to 1.5 W / (m·K). Thermal resistance is calculated using a thermal network model, where the thermal resistance between each cell is expressed as the reciprocal of the thermal conductivity coefficient multiplied by the conduction path length, then divided by the contact area. For example, the thermal resistance between adjacent cells ranges from 0.5 K / W to 2.0 K / W. The temperature field conduction equation is solved using the finite difference method, with a time step of 10 seconds and a spatial step of 5 mm. The heat flux density distribution obtained from the solution shows that during fast charging, the heat flux density in the central region of the battery cluster can reach 120 W / m². 2 The edge region has a power consumption of approximately 60 W / m. 2 .

[0098] Thermal stress calculations are based on the coefficient of thermal expansion and temperature differences. The linear coefficient of thermal expansion of the electrode material is approximately 1.6 × 10⁻⁶. -5 / °C. The stress distribution characteristics were obtained through finite element analysis. The results showed that the maximum stress inside the battery was concentrated in the contact area between the positive and negative electrodes, reaching a value of 18 MPa. The coupling between temperature distribution characteristics and stress distribution characteristics was determined using a weighted coefficient method to construct a temperature-stress coupling feature vector. The temperature-stress coupling feature vector contains three components: temperature gradient, stress gradient, and their product. The weighted coefficients were determined through training with historical battery performance data, with typical values ​​of [0.4, 0.3, 0.3].

[0099] The fused feature vector is constructed using feature concatenation and dimensionality reduction methods. The spatial temperature gradient (3×n×m×l), the temporal temperature change rate (n×m×l), and the temperature stress coupling feature (3×n×m×l) are concatenated with the original feature parameters (including voltage, current, SOC, etc., with a dimension of k) to obtain a high-dimensional feature vector. Principal component analysis is then used to reduce its dimensionality to 128 dimensions, retaining over 95% of the information.

[0100] The deep reinforcement learning algorithm is implemented using a dual deep Q-network framework. The state space is a fused feature vector, and the action space consists of adjusted values ​​for the predicted capacity decay rate and internal resistance growth rate, ranging from [-0.5%, 0.5%] to [-1%, 1%], discretely represented by 21 values. The reward function is designed based on the prediction error; the smaller the prediction error, the higher the reward. The network structure contains four fully connected layers with 256, 128, 64, and 32 neurons respectively, and the activation function is Leaky ReLU. During training, the batch size is set to 64, the learning rate to 0.001, the discount factor to 0.95, and the target network update frequency to 100 steps. The training data includes cyclic test data of 200 battery clusters under different operating conditions, each containing at least 500 charge-discharge cycles.

[0101] Modeling results of capacity decay indicate that, under standard operating conditions (25°C ambient temperature, 1C charge / discharge rate), the capacity decay of individual cells in the battery cluster follows a double exponential decay pattern, with a decay rate of approximately 0.15% / cycle for the first 100 cycles, decreasing to 0.08% / cycle in the later stages. The internal resistance growth exhibits an approximately linear trend, with a growth rate of approximately 0.2% / cycle. Temperature gradients have a significant impact on capacity decay; for every 1°C / cm increase in temperature gradient, the capacity decay rate increases by approximately 0.03% / cycle.

[0102] The initial capacity retention rate and initial internal resistance change rate are calculated based on the baseline values ​​of a new battery, where the capacity and internal resistance are defined as 100%. Using a capacity decay model to predict the capacity after n cycles, let Cn be the capacity retention rate; then, the capacity retention rate is Cn / C0 × 100%. Similarly, the internal resistance change rate is Rn / R0 × 100%, where Rn is the internal resistance after n cycles. Corrections for spatial temperature gradient and temperature stress coupling characteristics are made using a linear adjustment method, with correction coefficients determined through regression analysis. For example, in a battery cluster test, the initially predicted capacity retention rate after 100 cycles was 92%. After temperature gradient correction (correction coefficient -0.5% / (°C / cm)) and temperature stress coupling characteristic correction (correction coefficient -0.3% / MPa), the final capacity retention rate was corrected to 89.6%, close to the measured value of 90.1%.

[0103] In this embodiment, both spatial temperature gradient and temporal temperature change rate are considered simultaneously, enabling accurate modeling of the temperature distribution of individual battery cells. By coupling the temperature distribution characteristics with the stress distribution characteristics, the limitations of traditional separate temperature and stress analyses are overcome, achieving synergistic analysis of the temperature and stress fields. Through multi-physics coupling analysis, a more reliable method for assessing battery health status is provided, offering strong technical support for the safe and stable operation of energy storage systems. This significantly improves the accuracy and reliability of battery health status assessment and is of great significance for extending the service life of energy storage systems.

[0104] Figure 2 This is a thermal stress distribution diagram of the battery electrode material in the collaborative scheduling and control method of the virtual power plant energy storage system according to an embodiment of the present invention, based on a precise temperature gradient and the linear thermal expansion coefficient of the electrode material (1.6 × 10⁻⁶). -5 The thermal stress distribution was calculated by finite element analysis (°C), and the temperature-stress coupling eigenvector (weighting coefficients [0.4, 0.3, 0.3]) was used to effectively integrate the temperature distribution characteristics and stress distribution characteristics.

[0105] like Figure 2 As shown, the maximum thermal stress in the contact area between the positive and negative electrodes reaches 17.8 MPa (red area in the figure), the thermal stress in the middle area of ​​the electrode is 8.4 MPa (yellow area in the figure), and the lowest in the edge area is only 3.2 MPa (blue area in the figure). The non-uniform distribution is highly consistent with the actual internal structure and temperature distribution of the battery, and the prediction accuracy reaches 94.6%. The traditional model assumes uniform thermal expansion and predicts a maximum thermal stress of only 12.5 MPa, which is far lower than the measured value of 18 MPa. Moreover, it cannot capture local high-stress areas, and the prediction accuracy is only 69.4%. This technical solution can accurately predict the stress concentration phenomenon in hot spot areas during rapid charging and discharging, providing a key basis for assessing battery performance degradation and potential safety risks.

[0106] In one alternative implementation,

[0107] Based on rolling time domain prediction of system operating states for multiple future time periods, system state prediction values ​​are obtained. A power allocation strategy is then dynamically generated based on these system state prediction values ​​and the initial operating parameters, including:

[0108] A rolling prediction interval and a rolling update cycle are set. A deep learning algorithm is used to predict the load power, grid power, energy storage power and state of charge within the rolling prediction interval. The predicted values ​​of load power, grid power, energy storage power and state of charge for each time period are obtained and combined to obtain the predicted value of system state.

[0109] Based on the system state prediction value and the initial operating parameters, calculate the maximum allowable fluctuation range of grid power and energy storage power, the maximum allowable charging and discharging range of energy storage power, and the maximum allowable exchange range of grid power, and generate an initial power allocation scheme.

[0110] The initial power allocation scheme is applied to the grid power forecast and energy storage power forecast for each time period to ensure that the sum of the grid power and energy storage power for each time period is equal to the load power forecast for the corresponding time period. The energy storage power is within the maximum allowable charge and discharge range, and the grid power is within the maximum allowable exchange range, thus obtaining the power allocation strategy.

[0111] The rolling forecast interval is set to 4 hours, with a rolling update cycle of 15 minutes. Within each rolling forecast interval, a deep learning algorithm based on a Long Short-Term Memory (LSTM) network is used to predict the load power, grid power, energy storage power, and state of charge (SBC) for future periods. Historical data is used as input during the forecasting process, including load power, grid power, energy storage power, and SBC data recorded every 15 minutes over the past 24 hours. Through training the LSTM network, predictions are made for load power, grid power, energy storage power, and SBC every 15 minutes over the next 4 hours, resulting in 16 predicted values ​​for a total of 16 periods. For example, in one forecast, the predicted load power for the first 15-minute period is 120 kW, the predicted grid power is 80 kW, the predicted energy storage power is 40 kW, and the predicted SBC is 75%. These predicted values ​​are combined to form the system state prediction.

[0112] After obtaining the system state prediction values, the maximum allowable fluctuation range of grid power and energy storage power, the maximum allowable charging and discharging range of energy storage power, and the maximum allowable exchange range of grid power are calculated based on the system state prediction values ​​and initial operating parameters, and an initial power allocation scheme is generated. The initial operating parameters include a maximum allowable grid power of 150kW, a minimum allowable power of -50kW (negative values ​​indicate power feeding to the grid), a maximum charging power of 60kW for the energy storage system, a maximum discharging power of 60kW, an upper limit of 95% of the state of charge (SOC) and a lower limit of 15%, a current SOC of 75%, and an energy storage system capacity of 200kWh.

[0113] The maximum allowable fluctuation range of grid power and energy storage power is calculated. Considering grid stability requirements, the maximum rate of change of grid power between two adjacent time periods is set to 20kW / 15 minutes, and the maximum rate of change of energy storage power between two adjacent time periods is set to 30kW / 15 minutes. For example, if the current grid power is 80kW, then the grid power in the next 15-minute time period should be between 60kW and 100kW; if the current energy storage power is 40kW, then the energy storage power in the next 15-minute time period should be between 10kW and 70kW.

[0114] The maximum allowable charge / discharge range of energy storage power is determined based on the state of charge (SBC) and physical limitations of the energy storage system. For example, when the SBC is 75%, there is still 20% of capacity to charge before reaching the 95% limit, corresponding to 40kWh of energy. Within 15 minutes, the maximum chargeable power is 60kW (considering a charging efficiency of approximately 90%), therefore the actual maximum charging power is 54kW. The current dischargeable capacity is 60% (75% - 15%), corresponding to 120kWh of energy. Within 15 minutes, the maximum dischargeable power is 60kW (considering a discharge efficiency of approximately 95%), therefore the actual maximum discharge power is 57kW. Taking all factors into account, the maximum allowable charge / discharge range of energy storage power is [-57kW, 54kW], where negative values ​​represent discharging and positive values ​​represent charging.

[0115] The maximum allowable power exchange range of the power grid is determined according to the technical limitations of the power grid. The maximum allowable power exchange range of the power grid is [-50kW, 150kW].

[0116] Based on the above calculations, an initial power allocation scheme is generated. For the first 15-minute period, the predicted load power is 120kW, and the initial allocation scheme is: grid power 80kW, energy storage power 40kW (discharging). For the second 15-minute period, the predicted load power is 125kW, and the initial allocation scheme is: grid power 85kW, energy storage power 40kW (discharging). For the third 15-minute period, the predicted load power is 130kW, and the initial allocation scheme is: grid power 90kW, energy storage power 40kW (discharging). An initial power allocation scheme will be generated for each 15-minute period within the next 4 hours.

[0117] The initial power allocation scheme is applied to the grid power forecast and energy storage power forecast for each time period, and adjustments are made to ensure that three conditions are met: the sum of the grid power and energy storage power for each time period equals the load power forecast for the corresponding time period; the energy storage power is within the maximum allowable charge and discharge range; and the grid power is within the maximum allowable exchange range.

[0118] For example, if the predicted load power for the fourth 15-minute period is 140kW, the initial allocation scheme is 95kW grid power and 45kW energy storage power (discharge). However, since the grid power in the previous period was 90kW, the grid power in the fourth period can only reach a maximum of 110kW due to the maximum rate of change. Therefore, the grid power is adjusted to 110kW grid power and 30kW energy storage power (discharge).

[0119] After adjustments, the final power allocation strategy is obtained. After each rolling update cycle (15 minutes), the prediction and power allocation process is re-executed to achieve dynamic rolling optimization.

[0120] In this embodiment, rolling prediction of system status is achieved through deep learning algorithms. This not only accurately grasps the dynamic changes in load power, grid power, and energy storage system operating status, but also ensures the real-time performance and accuracy of predictions by setting rolling prediction intervals and update cycles. The multi-dimensional constraint design not only ensures the safety of system operation but also improves the feasibility of power allocation schemes. Through the organic combination of prediction and constraints, both the economy and stability of the system are guaranteed, providing effective technical support for the optimized scheduling of energy storage systems.

[0121] Figure 3 This is a flowchart illustrating the power allocation strategy generation process of the collaborative scheduling and control method for a virtual power plant energy storage system according to an embodiment of the present invention.

[0122] In one alternative implementation,

[0123] By employing mixed-integer linear programming and model predictive control methods, combined with the power allocation strategy, the optimal scheduling scheme for each time period is calculated. Taking into account grid fluctuations, load changes, and charging / discharging benefits, the optimal scheduling scheme is then corrected in real time, resulting in time-segmented scheduling instructions for the energy storage system, including:

[0124] Extract power time-series data to calculate power fluctuation sequence, extract load power time-series data to calculate load change sequence, construct the current power grid power and load power as the root node state vector, superimpose the power grid fluctuation sequence and load change sequence onto the root node state vector at preset time intervals to obtain child node state vectors at different times, construct the root node state vector and child node state vectors into a scene tree structure, and calculate the occurrence probability of each node in the scene tree structure and the transition probability between adjacent nodes;

[0125] Based on the scenario tree structure, establish a continuous variable vector containing energy storage power variables and state of charge variables, establish an integer variable vector containing charging state indicator variables and discharging state indicator variables, establish a power balance constraint condition that the sum of grid power and energy storage power equals the load power, and establish energy storage system operation constraints for the range of energy storage power and state of charge values.

[0126] The continuous variable vector and integer variable vector are linearized. Under the constraints of the power balance constraint and the energy storage system operation constraint, the energy storage power and charge / discharge state corresponding to the minimum cost are calculated to obtain the initial optimal solution. A prediction time window and a control time window are set. Within the prediction time window, the predicted values ​​of the grid power and load power for future periods are calculated based on the current grid power and load power. Within the control time window, the energy storage power and charge / discharge state for each period are calculated. The power allocation strategy is used as a constraint to optimize and iteratively calculate the energy storage power and charge / discharge state to obtain the optimal scheduling scheme for each period.

[0127] Calculate the deviation between the current grid power and load power and the grid power and load power in the sub-node state vector. Based on the deviation, calculate the scenario weight coefficient and power correction amount. Combine the grid fluctuation sequence, load change sequence and the charging and discharging benefits, and superimpose the power correction amount with the energy storage power in the optimal scheduling scheme to obtain the time-sharing scheduling instruction of the energy storage system.

[0128] The power grid data for the most recent 7 days is extracted from the historical database, differentially processed, and the power grid change value for each time period is calculated to form a power grid fluctuation sequence. For example, if the power of a power grid is 100kW, 102kW, 105kW, etc. in consecutive 24 hours, the corresponding power grid fluctuation sequence is 2kW, 3kW, etc. Similarly, the load power data for the most recent 7 days is extracted, and the load power change value for each time period is calculated to form a load change sequence. Assuming the load power is 95kW, 98kW, 103kW, etc., the load change sequence is 3kW, 5kW, etc.

[0129] The grid power Pg(t0) and load power Pl(t0) at the current time t0 are combined into the root node state vector S0 = [Pg(t0), Pl(t0)]. For example, if the current grid power is 100kW and the load power is 95kW, then the root node state vector S0 = [100kW, 95kW]. Based on a preset time interval Δt (e.g., 15 minutes), the grid fluctuation sequence and load change sequence are superimposed onto the root node state vector to obtain the child node state vectors. For example, the first child node state vector S1 = [100kW + 2kW, 95kW + 3kW] = [102kW, 98kW], and the second child node state vector S2 = [102kW + 3kW, 98kW + 5kW] = [105kW, 103kW]. The root node and all child nodes are constructed into a scene tree structure, with each node representing the system state at a given time. Based on statistical analysis of historical data, the occurrence probability of each node and the transition probability between adjacent nodes are calculated. For example, based on historical data analysis, the probability of the grid power changing from 100kW to 102kW is 0.7, and the probability of changing to 98kW is 0.3.

[0130] Based on the scene tree structure, a mixed-integer linear programming model is established. A continuous variable vector X = [Ps(1), Ps(2), ..., Ps(N), SOC(1), SOC(2), ..., SOC(N)] is defined, where Ps(i) represents the energy storage power in the i-th time period, and SOC(i) represents the state of charge in the i-th time period. An integer variable vector Y = [δc(1), δc(2), ..., δc(N), δd(1), δd(2), ..., δd(N)] is defined, where δc(i) and δd(i) are the charging state indicator and discharging state indicator variables in the i-th time period, respectively, and their values ​​are 0 or 1.

[0131] Establish the power balance constraint condition: For each time period i, Pg(i) + Ps(i) = Pl(i). Where Pg(i) is the grid power, Ps(i) is the energy storage power, and Pl(i) is the load power. For example, if the grid power is 100kW and the load power is 95kW at a certain moment, then the energy storage power Ps should be -5kW (the negative value indicates charging).

[0132] Establish the following operational constraints for the energy storage system: For each time period i, the energy storage power Ps(i) ranges from [-Ps_max, Ps_max], where Ps_max is the maximum power of the energy storage system, such as 100kW; the state of charge (SOC) ranges from [SOC_min, SOC_max], where SOC_min and SOC_max are the minimum and maximum states of charge, such as 20% and 90%, respectively. The charge / discharge state indicator variable satisfies δc(i) + δd(i) ≤ 1, indicating that the energy storage system cannot charge and discharge simultaneously at the same time.

[0133] Linearize continuous and integer variables to avoid nonlinear terms in the model. For example, express the charging and discharging power as Ps(i) = Pc(i) - Pd(i), where Pc(i) = δc(i) × Ps(i) represents the charging power, and Pd(i) = δd(i) × Ps(i) represents the discharging power. Linearize the product terms.

[0134] Under the constraints of power balance and energy storage system operation, calculate the energy storage power and state of charge / discharge that minimizes the total operating cost. The total cost includes grid purchase cost and energy storage system loss cost. For example, assuming an electricity price of 0.8 yuan / kWh and an energy storage system charge / discharge efficiency of 90%, the total cost of charging 1kWh of energy storage in a certain period is 0.8 × 1 ÷ 0.9 = 0.89 yuan. By solving the optimization problem, an initial optimal solution is obtained, including the energy storage power and state of charge / discharge for each period.

[0135] Set the prediction time window \(T_p\) (e.g., 24 hours) and the control time window \(T_c\) (e.g., 4 hours), where \(T_c < T_p\). Based on the current grid power and load power, use the autoregressive integrated moving average model to calculate the predicted values of grid power and load power within the future \(T_p\) period. For example, if the current grid power is 100 kW, the predicted values of grid power per hour within the next 24 hours calculated according to the prediction model are 102 kW, 105 kW, etc.

[0136] Within the control time window \(T_c\), combined with the power allocation strategy, calculate the energy storage power and charge-discharge state for each period. The power allocation strategy takes into account the characteristics of different types of energy storage devices. For example, battery energy storage is suitable for handling medium- and long-term fluctuations, and flywheel energy storage is suitable for handling short-term fluctuations. For a hybrid energy storage system composed of 10% battery energy storage and 90% flywheel energy storage, the fluctuating power is allocated according to the frequency characteristics, with the high-frequency part handled by flywheel energy storage and the low-frequency part handled by battery energy storage.

[0137] Through iterative optimization calculations, obtain the optimal scheduling plans for each period. Represent these plans as \([P_s^*(1), P_s^*(2),..., P_s^*(T_c)]\) and \([\delta_c^*(1), \delta_c^*(2),..., \delta_c^*(T_c), \delta_d^*(1), \delta_d^*(2),..., \delta_d^*(T_c)]\).

[0138] Real-time monitor the current grid power \(P_{g\_real}\) and load power \(P_{l\_real}\), and calculate the deviations from the predicted values \(\Delta P_g = P_{g\_real}-P_{g\_pred}\) and \(\Delta P_l = P_{l\_real}-P_{l\_pred}\). For example, if the predicted grid power is 100 kW and the actual value is 105 kW, then the deviation \(\Delta P_g = 5\) kW. Based on the deviation values, calculate the scenario weight coefficient \(w_i\) and the power correction amount \(\Delta P_s\). For example, if the deviation is large, increase the weight of the corresponding scenario and adjust the power correction amount accordingly.

[0139] Combined with the grid fluctuation sequence, load change sequence, and charge-discharge benefits, superimpose the power correction amount \(\Delta P_s\) on the energy storage power \(P_s^*\) in the optimal scheduling plan to obtain the corrected energy storage power \(P_{s\_adj}=P_s^*+\Delta P_s\). For example, if the optimal energy storage power for a certain period is -10 kW and the power correction amount is 2 kW, then the corrected energy storage power is -8 kW.

[0140] According to the corrected energy storage power and charge-discharge state, generate the time-segmented scheduling instructions for the energy storage system, including the charge-discharge power values and operating modes for each period. The time-segmented scheduling instructions are sent to the energy storage device actuator through the control system to achieve the optimal operation of the energy storage system.

[0141] In this embodiment, by extracting the temporal variation characteristics of grid power and load power, a dynamic evolution path from the root node to the child node is established. This not only captures the stochastic characteristics of system operation but also achieves scientific prediction through the calculation of occurrence and transition probabilities. By establishing power balance constraints and operational constraints, the feasibility of the optimization results is ensured. A rolling optimization strategy with dual time windows is adopted. The long-term change trend of the system is captured by predicting the time window, and the precise regulation of the energy storage system is achieved by controlling the time window. This not only improves the operating efficiency of the energy storage system but also enhances the system's adaptability to changes in the external environment, providing a strong guarantee for the economical and efficient operation of the energy storage system.

[0142] Figure 4 This is a comparison chart of the power grid power balance performance of the collaborative scheduling and control method of the virtual power plant energy storage system in an embodiment of the present invention, showing the difference in power fluctuation control between the present technical solution and the traditional PID control algorithm and fixed threshold control method.

[0143] This technical solution achieves minimal grid power fluctuation, with a fluctuation range of 7-15kW. The traditional PID control algorithm has a fluctuation range of 18-25kW, while the fixed threshold control method has a fluctuation range of 30-38kW. At the start of the test (0 hours), the power fluctuations of the three methods were 15kW, 25kW, and 35kW, respectively. By the 8th hour, the power fluctuations of the three methods had decreased to 7kW, 18kW, and 30kW, respectively. By the 16th hour, the power fluctuations were 11kW, 23kW, and 38kW, respectively. By the 24th hour, the fluctuations of the three methods were 10kW, 21kW, and 37kW, respectively. Finally, at the 32nd hour (the end of the test), the fluctuations of the three methods were 12kW, 22kW, and 38kW, respectively.

[0144] Throughout the entire testing period, the average power fluctuation of this technical solution was approximately 11kW, which is about 50% lower than the average fluctuation of 22kW of the traditional PID control algorithm and about 69% lower than the average fluctuation of 35kW of the fixed threshold control method. The advantages of this technical solution are particularly evident during the high-load period from the 10th to the 14th hour, with the power fluctuation remaining at around 9kW, while the traditional PID control and fixed threshold control reached 19kW and 34kW, respectively.

[0145] This technical solution takes into account grid fluctuations, load changes, and charging and discharging benefits. Through a scenario tree structure and a real-time correction mechanism, it dynamically adjusts the charging and discharging strategy of the energy storage system. After a long period of operation (28-32 hours), the power fluctuation of this technical solution remains stable at around 12kW, while the other two methods rise to 22kW and 38kW respectively, demonstrating the advantages of this solution in long-term stable control.

[0146] In one alternative implementation,

[0147] According to the time-segmented scheduling instructions, charging and discharging operations are performed, and the operating parameters of the energy storage system are monitored in real time. When the operating parameters exceed a preset safety threshold, the charging and discharging power is adjusted according to a preset emergency control strategy, including:

[0148] Receive time-segmented scheduling instructions, and control the energy storage system to perform charging and discharging operations according to the energy storage power and charging / discharging status in the time-segmented scheduling instructions;

[0149] Monitor the real-time operating parameters of the energy storage system and compare them with the corresponding preset safety thresholds;

[0150] When any parameter exceeds the corresponding preset safety threshold, the charging power or discharging power is reduced according to the power adjustment rules in the preset emergency control strategy until the operating parameter returns to the preset safety threshold range.

[0151] The system receives time-segmented dispatch instructions from the power grid dispatch center via a communication interface. These instructions include the energy storage power value and charging / discharging status for each time period. For example, the dispatch instruction might specify discharging at 500kW during peak grid load periods (e.g., 18:00-21:00) and charging at 400kW during off-peak periods (e.g., 02:00-05:00). The energy storage system's controller parses the instructions and determines the appropriate charging / discharging operation based on the current system clock. The controller then sends corresponding control signals to the energy storage units via the power conversion system (PCS) to control the charging / discharging status and power output of the energy storage units.

[0152] While performing charging and discharging operations, the energy storage system continuously monitors various operating parameters, including but not limited to: battery temperature, battery voltage, battery current, state of charge (SOC), internal impedance of the energy storage unit, and power conversion efficiency. The monitoring frequency is set according to the importance of the parameters; critical parameters such as battery temperature can be monitored once per second, while relatively stable parameters such as internal impedance can be monitored once per minute. The real-time collected parameter data is transmitted to the controller for processing and analysis via a data acquisition system.

[0153] The controller compares the various parameters monitored in real time with preset safety thresholds. These safety thresholds are determined based on the energy storage system equipment specifications and actual operating experience. For example, for lithium-ion battery energy storage systems, the safety threshold for individual battery cell temperature can be set to -20℃ to 55℃; exceeding this range will trigger emergency control. The safety threshold for individual battery cell voltage can be set to 2.5V to 4.2V. The safe operating range for state of charge can be set to 10% to 90% to avoid overcharging and over-discharging. The safe threshold for battery charging current can be set to no more than 0.5C (i.e., 0.5 times the battery's rated capacity). The safe threshold for discharging current can be set to no more than 1C.

[0154] When any parameter is detected to exceed its corresponding preset safety threshold, the controller will activate an emergency control strategy. This strategy includes power adjustment rules for different abnormal parameter conditions. The controller dynamically adjusts the charging and discharging power based on the type of abnormal parameter and the degree of deviation from the safety threshold. For example, when the battery temperature reaches 50°C (close to the upper limit of 55°C), the controller will linearly reduce the charging power from the original 400kW to 200kW according to preset rules to reduce heat generation; if the temperature continues to rise to 52°C, the charging power will be further reduced to 100kW; if it reaches 54°C, charging will be completely stopped.

[0155] Different emergency control strategies are set for different parameters. In cases of abnormal state of charge (SOC), when the SOC reaches 85% (close to the upper limit of 90%), the charging power will automatically decrease to 50% of the original power; when the SOC reaches 88%, the charging power will further decrease to 20% of the original power; and when the SOC reaches 89%, the charging power will decrease to the minimum value or charging will stop completely. Similarly, when the SOC drops to 15% (close to the lower limit of 10%), the discharging power will decrease to 50% of the original power; when the SOC drops to 12%, the discharging power will decrease to 20% of the original power; and when the SOC drops to 11%, the discharging power will decrease to the minimum value or discharging will stop completely.

[0156] In case of abnormal battery voltage, when the voltage of a single battery cell reaches 4.1V (close to the upper limit of 4.2V), the charging power will be reduced to 70% of the original power; when the voltage reaches 4.15V, the charging power will be reduced to 30% of the original power; when the voltage reaches 4.18V, the battery cell will stop charging. When the voltage of a single battery cell drops to 2.7V (close to the lower limit of 2.5V), the discharge power will be reduced to 70% of the original power; when the voltage drops to 2.6V, the discharge power will be reduced to 30% of the original power; when the voltage drops to 2.55V, the battery cell will stop discharging.

[0157] The emergency control strategy also includes a recovery mechanism, which gradually restores the charging and discharging power once the abnormal parameters return to a certain margin within the safety threshold range. For example, when the battery temperature drops from the abnormally high temperature of 50°C to 45°C (10°C below the safety threshold of 55°C), the charging power is restored to 50% of the original power; when the temperature further drops to 40°C, the charging power is fully restored to the original power of 400kW.

[0158] In this embodiment, by executing the energy storage power and charge / discharge status control requirements in the time-sharing scheduling command, precise scheduling control of the energy storage system is achieved, laying the foundation for the economical operation of the system. Real-time monitoring based on thresholds can detect problems in the early stages of abnormal situations, preventing further expansion of the fault. The flexible power adjustment strategy maintains the continuous operation of the system to the greatest extent while ensuring system safety, thereby improving the reliability and utilization efficiency of the energy storage system.

[0159] A second aspect of the present invention provides a collaborative scheduling and control system for a virtual power plant energy storage system, comprising:

[0160] The first unit is used to acquire grid dispatch data within a preset time period and calculate the charging and discharging benefits of the energy storage system under the condition of meeting system constraints.

[0161] The second unit is used to collect real-time operating parameters of individual battery clusters in the energy storage system, analyze the real-time operating parameters based on deep reinforcement learning methods, obtain the performance degradation law of individual battery clusters, and determine the initial operating parameters of each individual battery cluster based on the performance degradation law.

[0162] The third unit is used to predict the system operating status for multiple future time periods based on the rolling time domain to obtain the system status prediction value. Based on the system status prediction value and the initial operating parameters, a power allocation strategy is dynamically generated. By using mixed integer linear programming and model predictive control methods, combined with the power allocation strategy, the optimal scheduling scheme for each time period is calculated. Taking into account grid fluctuations, load changes and charging and discharging benefits, the optimal scheduling scheme is corrected in real time to obtain the time-sharing scheduling instructions for the energy storage system.

[0163] The fourth unit is used to perform charging and discharging operations according to the time-sharing scheduling instructions and monitor the operating parameters of the energy storage system in real time. When the operating parameters exceed the preset safety threshold, the charging and discharging power is adjusted according to the preset emergency control strategy.

[0164] A third aspect of the present invention provides an electronic device, comprising:

[0165] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0166] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0167] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for collaborative scheduling and control of a virtual power plant energy storage system, characterized in that, include: Obtain grid dispatch data within a preset time period and calculate the charging and discharging benefits of the energy storage system under system constraints. Real-time operating parameters of individual battery clusters in the energy storage system are collected, and the real-time operating parameters are analyzed based on deep reinforcement learning methods to obtain the performance degradation law of individual battery clusters. The initial operating parameters of each individual battery cluster are determined based on the performance degradation law. Based on the rolling time domain, the system operating status of multiple future time periods is predicted to obtain the system status prediction value. The power allocation strategy is dynamically generated according to the system status prediction value and the initial operating parameters. The optimal scheduling scheme for each time period is calculated by combining the power allocation strategy with mixed integer linear programming and model predictive control methods. The optimal scheduling scheme is corrected in real time by comprehensively considering grid fluctuations, load changes and charging and discharging benefits to obtain the time-sharing scheduling instructions of the energy storage system. The system executes charging and discharging operations according to the time-segmented scheduling instructions and monitors the operating parameters of the energy storage system in real time. When the operating parameters exceed the preset safety threshold, the charging and discharging power is adjusted according to the preset emergency control strategy. By employing mixed-integer linear programming and model predictive control methods, combined with the power allocation strategy, the optimal scheduling scheme for each time period is calculated. Taking into account grid fluctuations, load changes, and charging / discharging benefits, the optimal scheduling scheme is then corrected in real time, resulting in time-segmented scheduling instructions for the energy storage system, including: Extract power time-series data to calculate power fluctuation sequence, extract load power time-series data to calculate load change sequence, construct the current power grid power and load power as the root node state vector, superimpose the power grid fluctuation sequence and load change sequence onto the root node state vector at preset time intervals to obtain child node state vectors at different times, construct the root node state vector and child node state vectors into a scene tree structure, and calculate the occurrence probability of each node in the scene tree structure and the transition probability between adjacent nodes; Based on the scenario tree structure, establish a continuous variable vector containing energy storage power variables and state of charge variables, establish an integer variable vector containing charging state indicator variables and discharging state indicator variables, establish a power balance constraint condition that the sum of grid power and energy storage power equals the load power, and establish energy storage system operation constraints for the range of energy storage power and state of charge values. The continuous variable vector and integer variable vector are linearized. Under the constraints of the power balance constraint and the energy storage system operation constraint, the energy storage power and charge / discharge state corresponding to the minimum cost are calculated to obtain the initial optimal solution. A prediction time window and a control time window are set. Within the prediction time window, the predicted values ​​of the grid power and load power for future periods are calculated based on the current grid power and load power. Within the control time window, the energy storage power and charge / discharge state for each period are calculated. The power allocation strategy is used as a constraint to optimize and iteratively calculate the energy storage power and charge / discharge state to obtain the optimal scheduling scheme for each period. Calculate the deviation between the current grid power and load power and the grid power and load power in the sub-node state vector. Based on the deviation, calculate the scenario weight coefficient and power correction amount. Combine the grid fluctuation sequence, load change sequence and the charging and discharging benefits, and superimpose the power correction amount with the energy storage power in the optimal scheduling scheme to obtain the time-sharing scheduling instruction of the energy storage system.

2. The method according to claim 1, characterized in that, Obtain grid dispatch data within a preset time period and calculate the charging and discharging benefits of the energy storage system under system constraints, including: Obtain grid load demand information and electricity price information within a preset time period; A time-series analysis of the power grid load demand information and electricity price information is performed to obtain the charging and discharging period division results, which include high load and high electricity price periods and low load and low electricity price periods; Based on the results of the charging and discharging time period division, and combined with the power and capacity constraints of the energy storage system, charging operations are performed during low-load and low-electricity-price periods, and discharging operations are performed during high-load and high-electricity-price periods, to calculate the charging and discharging revenue of the energy storage system.

3. The method according to claim 1, characterized in that, Real-time operating parameters of individual battery cells in the energy storage system are collected. These parameters are then analyzed using deep reinforcement learning to determine the performance degradation patterns of the individual battery cells. Based on these performance degradation patterns, the initial operating parameters of each individual battery cell are determined, including: Real-time operating parameters of individual battery cells are collected, and outliers are removed using a three-standard-deviation criterion. The outliers are then standardized by maximum and minimum values. Wavelet transform is used to extract time-domain features from the standardized real-time operating parameters to obtain the characteristic parameters of individual battery cells. The real-time operating parameters include terminal voltage parameters, open-circuit voltage parameters, battery state of charge, cycle number, surface temperature parameters, and internal temperature distribution parameters. The feature parameters are analyzed based on a deep reinforcement learning algorithm to obtain the capacity decay law and internal resistance growth law of individual battery clusters, and the capacity retention rate and internal resistance change rate of individual battery clusters are calculated. The health status of individual battery clusters is determined based on the capacity retention rate and internal resistance change rate. The charge / discharge depth limit is determined based on the health status and the real-time operating parameters. The power of each battery cluster cell is allocated based on the charge / discharge depth limit to obtain the initial operating parameters of each battery cluster cell.

4. The method according to claim 3, characterized in that, The feature parameters are analyzed using a deep reinforcement learning algorithm to obtain the capacity decay law and internal resistance growth law of individual battery cluster cells, and the capacity retention rate and internal resistance change rate of individual battery cluster cells are calculated, including: A temperature field gradient conduction network is established based on the aforementioned characteristic parameters. The spatial temperature gradient and time-varying temperature change rate of the surface temperature parameter and the internal temperature distribution parameter among the characteristic parameters are calculated. The temperature distribution characteristics of individual cells in the battery cluster are determined based on the spatial temperature gradient and the time-varying temperature change rate. A thermal conduction matrix is ​​established based on the physical positional relationship of the individual cells in the battery cluster. The thermal resistance between cells is calculated and a temperature field conduction equation is established to obtain the thermal flow conduction characteristics of the individual cells in the battery cluster. The thermal stress is calculated based on the thermal flow conduction characteristics and the stress distribution characteristics of the electrode material are determined. The stress distribution characteristics are coupled with the temperature distribution characteristics to obtain the temperature-stress coupling characteristics. The spatial temperature gradient, time-temperature change rate, and temperature stress coupling features are fused with the feature parameters to obtain a fused feature vector. Based on the deep reinforcement learning algorithm, the fused feature vector is analyzed to establish a temperature field capacity decay coupling model and a temperature field internal resistance growth coupling model, thereby obtaining the capacity decay law and internal resistance growth law of the battery cluster cells. The initial capacity retention rate and initial internal resistance change rate of the individual cells in the battery cluster are calculated based on the capacity decay law and internal resistance growth law. The initial capacity retention rate and initial internal resistance change rate are then corrected by the spatial temperature gradient and temperature stress coupling characteristics to obtain the capacity retention rate and internal resistance change rate of the individual cells in the battery cluster.

5. The method according to claim 1, characterized in that, Based on rolling time domain prediction of system operating states for multiple future time periods, system state prediction values ​​are obtained. A power allocation strategy is then dynamically generated based on these system state prediction values ​​and the initial operating parameters, including: A rolling prediction interval and a rolling update cycle are set. A deep learning algorithm is used to predict the load power, grid power, energy storage power and state of charge within the rolling prediction interval. The predicted values ​​of load power, grid power, energy storage power and state of charge for each time period are obtained and combined to obtain the predicted value of system state. Based on the system state prediction value and the initial operating parameters, calculate the maximum allowable fluctuation range of grid power and energy storage power, the maximum allowable charging and discharging range of energy storage power, and the maximum allowable exchange range of grid power, and generate an initial power allocation scheme. The initial power allocation scheme is applied to the grid power forecast and energy storage power forecast for each time period to ensure that the sum of the grid power and energy storage power for each time period is equal to the load power forecast for the corresponding time period. The energy storage power is within the maximum allowable charge and discharge range, and the grid power is within the maximum allowable exchange range, thus obtaining the power allocation strategy.

6. The method according to claim 1, characterized in that, According to the time-segmented scheduling instructions, charging and discharging operations are performed, and the operating parameters of the energy storage system are monitored in real time. When the operating parameters exceed a preset safety threshold, the charging and discharging power is adjusted according to a preset emergency control strategy, including: Receive time-segmented scheduling instructions, and control the energy storage system to perform charging and discharging operations according to the energy storage power and charging / discharging status in the time-segmented scheduling instructions; Monitor the real-time operating parameters of the energy storage system and compare them with the corresponding preset safety thresholds; When any parameter exceeds the corresponding preset safety threshold, the charging power or discharging power is reduced according to the power adjustment rules in the preset emergency control strategy until the operating parameter returns to the preset safety threshold range.

7. A virtual power plant energy storage system collaborative scheduling and control system, used to implement the method of any one of claims 1-6, characterized in that, include: The first unit is used to acquire grid dispatch data within a preset time period and calculate the charging and discharging benefits of the energy storage system under the condition of meeting system constraints. The second unit is used to collect real-time operating parameters of individual battery clusters in the energy storage system, analyze the real-time operating parameters based on deep reinforcement learning methods, obtain the performance degradation law of individual battery clusters, and determine the initial operating parameters of each individual battery cluster based on the performance degradation law. The third unit is used to predict the system operating status for multiple future time periods based on the rolling time domain to obtain the system status prediction value. Based on the system status prediction value and the initial operating parameters, a power allocation strategy is dynamically generated. By using mixed integer linear programming and model predictive control methods, combined with the power allocation strategy, the optimal scheduling scheme for each time period is calculated. Taking into account grid fluctuations, load changes and charging and discharging benefits, the optimal scheduling scheme is corrected in real time to obtain the time-sharing scheduling instructions for the energy storage system. The fourth unit is used to perform charging and discharging operations according to the time-sharing scheduling instructions and monitor the operating parameters of the energy storage system in real time. When the operating parameters exceed the preset safety threshold, the charging and discharging power is adjusted according to the preset emergency control strategy.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images