Capacity recovery method based on full life cycle of lithium ion battery

Abstract

The invention relates to the technical field of energy storage battery management, and discloses a capacity recovery method based on the full life cycle of a lithium ion battery. According to the method, the uncertainty of power grid demand and environment temperature is modeled through a scene tree generation algorithm, a low-sensitivity temperature region is identified through sensitivity analysis to generate a robust temperature parameter candidate set, and a recovery time window is identified based on a battery aging characteristic prediction sequence. Constructing a dual-objective optimization problem of power grid auxiliary service income and battery full life cycle value, solving a conditional optimal decision scheme under each scene through a dynamic programming algorithm, selecting a decision scheme with a maximum worst condition target value by using a robust optimization criterion, extracting an execution instruction, outputting the execution instruction to a control system, and performing power grid auxiliary service income and battery full life cycle value optimization. And the decision scheme is adjusted in real time through a rolling optimization mechanism. According to the method, the robustness of balancing economic benefits and battery health in an uncertain environment by a capacity recovery decision is improved.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery management technology, and more specifically, to a capacity recovery method based on the entire life cycle of a lithium-ion battery. Background Technology

[0002] During operation, energy storage power stations need to perform capacity restoration operations to maintain battery performance, and at the same time, they need to respond to grid dispatch instructions to provide ancillary services and obtain economic benefits. Since energy storage power stations are deployed in outdoor environments, capacity restoration operations are affected by ambient temperature fluctuations, and the demand for grid ancillary services and environmental conditions are subject to predictive uncertainties in the future time window.

[0003] Traditional capacity restoration scheduling methods suffer from two main technical flaws. First, they only consider the aging trend of the batteries themselves when scheduling restoration times, failing to incorporate the temporal distribution of grid ancillary service demand and price incentives into the decision-making process. This leads to economic losses when the fixed-schedule capacity restoration time conflicts with high-value grid service periods, while prioritizing grid demand entirely may miss the optimal execution opportunity for battery capacity restoration, resulting in rapid battery performance degradation. Second, they rely on deterministic forecasts for optimization decisions, treating ambient temperature and grid price within a future time window as single, fixed values ​​without quantitatively modeling the forecast error range and actual fluctuation amplitude. When the actual ambient temperature deviates from the forecast due to meteorological changes, the capacity restoration temperature parameters optimized based on the forecast temperature cannot be achieved in the actual environment. Furthermore, the point-optimized temperature parameters used in traditional methods are calibrated in a laboratory constant-temperature environment. When environmental disturbances cause the battery temperature to deviate from the target temperature value, the capacity restoration efficiency drops significantly. Simultaneously, grid price fluctuations alter the service revenue structure, rendering the original scheduling scheme economically ineffective.

[0004] The aforementioned problems result in traditional capacity recovery and scheduling methods having high vulnerability and poor fault tolerance in decision-making schemes under uncertain environments, making it impossible to effectively balance the dual objectives of economic benefits of grid ancillary services and battery life cycle health management. Summary of the Invention

[0005] This invention provides a capacity recovery method based on the entire life cycle of lithium-ion batteries, which solves the technical problems in related technologies that the capacity recovery decision does not consider the coordinated optimization of grid ancillary service needs and battery health, and that the decision scheme based on deterministic prediction value optimization is highly vulnerable to uncertain environments.

[0006] This invention discloses a capacity recovery method based on the entire life cycle of lithium-ion batteries, applied to the battery management system of energy storage power stations, including the following steps: Obtain forecast data on power grid ancillary service demand and ambient temperature within a future time window, along with the corresponding uncertainty range, and organize them into a set of uncertainty parameters. Capacity recovery efficiency data of the battery at multiple temperature points is obtained, and a continuous function curve of capacity recovery efficiency versus temperature is fitted. Sensitivity analysis is performed on the function curve, and the gradient of capacity recovery efficiency with temperature deviation at each temperature point is calculated. Temperature ranges with absolute gradient values ​​less than the sensitivity threshold are identified as low-sensitivity regions. In the low-sensitivity regions, robustness evaluation indexes are used for evaluation. Temperature points with robustness evaluation index values ​​higher than the robustness threshold are selected to generate a robust temperature parameter candidate set. Obtain the battery aging feature prediction sequence, identify the necessary execution time window and the preferred execution time window for capacity recovery operation based on the aging feature prediction sequence, assign a first flexibility weight to the necessary execution time window, assign a second flexibility weight to the preferred execution time window, and generate a recovery time flexibility label. Using a scene tree generation algorithm, a scene set is generated based on an uncertainty parameter set and a Monte Carlo sampling method. A scene reduction algorithm is used to filter the scene set to obtain a representative scene set. A probability weight is assigned to each representative scene, and the scenes and their probability weights are organized into a tree structure to generate a scene tree structure. A joint decision objective function is constructed, which includes two sub-objectives: maximizing the revenue of grid ancillary services and maximizing the value of the battery throughout its life cycle. The two sub-objectives are weighted and fused. The decision variables include the scheduling decisions for each time period and the temperature parameters used during capacity recovery. The constraints include hard constraints on recovery and the requirement that the temperature parameters must be within the robust temperature parameter candidate set, thus generating an economic-health bi-objective optimization problem. In each branch scenario of the scenario tree structure, the dynamic programming algorithm is used to solve the bi-objective optimization problem. Starting from the end of the time window, the optimal decision of each stage is solved in reverse to generate the set of conditionally optimal decision schemes for each scenario. For each scheme in the set of optimal decision schemes, calculate the worst-case objective value of each scheme under all scenarios, and select the scheme with the largest worst-case objective value as the robust optimal decision scheme. The execution instructions for the current period are extracted from the robust optimal decision scheme. The execution instructions include scheduling decisions and robust temperature control parameters. The execution instructions are then output to the energy storage system controller and the battery thermal management system. Real-time monitoring of actual values ​​of grid demand, battery status, and ambient temperature; when the deviation between actual and predicted values ​​exceeds the deviation threshold, updating the probability weights of each scenario in the scenario tree structure, triggering rolling optimization, and recalculating the scheduling scheme for subsequent time periods. The robustness evaluation index is obtained by calculating the mean and standard deviation of capacity recovery efficiency in the low-sensitivity region, standardizing the mean and standard deviation of capacity recovery efficiency, and calculating the difference between the product of the standardized mean capacity recovery efficiency and the risk aversion coefficient and the standardized standard deviation of capacity recovery efficiency. The sensitivity threshold is determined based on the temperature control accuracy of the battery thermal management system and is calculated as the ratio of the maximum allowable rate of change of capacity recovery efficiency to the temperature control error.

[0007] This invention discloses a capacity recovery system based on the entire life cycle of a lithium-ion battery, comprising: The uncertainty parameter set generation module is used to obtain power grid ancillary service demand forecast data, ambient temperature forecast data and corresponding uncertainty range within a future time window and generate an uncertainty parameter set. The robust temperature parameter generation module is used to acquire capacity recovery efficiency data of the battery at multiple temperature points and generate a continuous function curve of capacity recovery efficiency versus temperature. It performs sensitivity analysis on the function curve to identify low-sensitivity regions and evaluates and generates a robust temperature parameter candidate set through robustness evaluation indicators. The recovery time flexibility analysis module is used to obtain the battery aging characteristic prediction sequence and identify the necessary execution time window and the preferred execution time window for capacity recovery operation, and to assign flexibility weights to different time windows to generate recovery time flexibility labels. The scene tree generation module is used to generate a scene set based on an uncertainty parameter set using the Monte Carlo sampling method, to filter a representative scene set using a scene reduction algorithm, to assign probability weights to each representative scene and to generate a scene tree structure. The dual-objective optimization problem construction module is used to construct a joint decision objective function that includes maximizing the revenue of grid ancillary services and maximizing the value of the battery throughout its entire life cycle. It weights and fuses the two sub-objectives and sets constraints to generate an economic-health dual-objective optimization problem. The conditionally optimal decision-making module is used to solve the bi-objective optimization problem in each branch of the scene tree structure using dynamic programming algorithm and generate a set of conditionally optimal decision schemes for each scene. The robust optimization decision selection module is used to calculate the worst-case objective value of each scheme in the set of optimal decision schemes under all scenarios and select the scheme with the largest worst-case objective value as the robust optimal decision scheme. The execution instruction generation module is used to extract the execution instructions for the current period from the robust optimal decision scheme and output them to the energy storage system controller and the battery thermal management system. The rolling optimization trigger module is used to monitor the actual values ​​of grid demand, battery status and ambient temperature in real time. When the deviation between the actual value and the predicted value exceeds the deviation threshold, the probability weight of each scenario in the scenario tree structure is updated and rolling optimization is triggered.

[0008] Furthermore, the step of acquiring battery capacity recovery efficiency data at multiple temperature points and fitting a continuous function curve of capacity recovery efficiency versus temperature includes: stabilizing the battery temperature at different temperature points by controlling the battery thermal management system; applying a capacity recovery operation to the battery at each temperature point and measuring the capacity recovery efficiency; employing a polynomial least squares fitting method, taking discrete temperature points and corresponding capacity recovery efficiency data as input, fitting a function in the form of a polynomial relationship between capacity recovery efficiency and temperature, with the polynomial order ranging from two to four; determining the polynomial coefficients by minimizing the sum of squared fitting errors to obtain a continuous function curve of capacity recovery efficiency versus temperature.

[0009] Furthermore, the sensitivity analysis of the function curve and the calculation of the gradient of capacity recovery efficiency with respect to temperature deviation at each temperature point includes: analytically differentiating the fitted continuous function curve of capacity recovery efficiency versus temperature; for a polynomial function, obtaining the gradient function by differentiating each term; calculating the gradient value at each temperature point, identifying temperature ranges where the absolute value of the gradient is less than the sensitivity threshold, and marking them as low-sensitivity regions; the sensitivity threshold is calculated by setting the maximum allowable rate of change of capacity recovery efficiency, dividing the maximum rate of change of capacity recovery efficiency by the temperature control error of the battery thermal management system, wherein the maximum rate of change of capacity recovery efficiency ranges from three percent to eight percent.

[0010] Further, the evaluation using robustness evaluation indicators in the low-sensitivity region includes: calculating the mean and standard deviation of capacity recovery efficiency within the low-sensitivity region; standardizing the mean and standard deviation of capacity recovery efficiency to eliminate the influence of dimensions; evaluating using robustness evaluation indicators, which are calculated as the standardized mean capacity recovery efficiency minus the product of the risk aversion coefficient and the standardized standard deviation of capacity recovery efficiency, with the risk aversion coefficient ranging from 0.5 to 2; selecting temperature points where the robustness evaluation indicator value is higher than the robustness threshold to generate a candidate set of robust temperature parameters, wherein the robustness threshold is calculated as the standardized minimum allowable capacity recovery efficiency minus the product of the risk aversion coefficient and the standardized maximum allowable standard deviation of capacity recovery efficiency, and the standardized maximum allowable standard deviation of capacity recovery efficiency is determined based on the maximum value of the standard deviation of capacity recovery efficiency within the low-sensitivity region.

[0011] Further, the step of acquiring the battery aging characteristic prediction sequence and identifying the necessary and preferred execution time windows for capacity recovery operations based on the aging characteristic prediction sequence includes: acquiring the battery aging characteristic prediction sequence, where aging characteristics include capacity decay rate, internal resistance growth rate, and polarization voltage change rate; the aging characteristic prediction sequence is generated using a time series prediction model based on historical monitoring data from the battery management system; and based on the aging characteristic prediction sequence, analyzing the time points when the capacity decay rate exceeds a decay threshold and the time points when the internal resistance growth rate exceeds an internal resistance threshold, and identifying the necessary execution time windows for capacity recovery operations, wherein the decay threshold value ranges from [value missing]. The battery capacity decays by 15% to 25% relative to the initial capacity. The internal resistance growth rate threshold ranges from 30% to 50% relative to the initial internal resistance growth rate. Simultaneously, the time point when the capacity decay rate approaches the decay threshold is analyzed to identify the preferred execution time window for capacity recovery. A first flexibility weight of 1 is assigned to the mandatory execution time window, and a second flexibility weight of less than 1 is assigned to the preferred execution time window. The value of the second flexibility weight ranges from 0.5 to 0.9. The smaller the difference between the capacity decay rate and the decay threshold, the larger the value of the second flexibility weight, generating a recovery time flexibility label.

[0012] Furthermore, the process of generating a scene set using the Monte Carlo sampling method and then filtering the scene set using a scene reduction algorithm to obtain a representative scene set includes: before sampling, performing mean normalization on the ambient temperature and grid price based on the range, mapping parameters with different dimensions and numerical ranges to the same numerical interval; randomly sampling based on the probability distribution of the normalized ambient temperature and grid price in the uncertainty parameter set, generating one scene per sampling, with each scene corresponding to a specific combination of uncertain parameters, including the ambient temperature sequence and grid price sequence under that scene; and filtering the scene set using a scene reduction algorithm. The similarity between scenes is calculated in the normalized parameter space. Representative scenes with large probability weights and significant differences are retained. Similar scenes are merged and their probability weights are accumulated. The similarity between scenes is calculated using Euclidean distance. The similarity between two scenes is obtained by taking the square root of the sum of the squares of the normalized temperature difference and the squares of the normalized price difference at each time period. When the similarity is less than the similarity threshold, the scenes are judged to be similar. The similarity threshold ranges from 5% to 15% of the diagonal length of the normalized parameter space. After the filtering is completed, the parameters of the representative scenes are denormalized to restore the temperature and price values ​​to their original dimensions.

[0013] Further, the construction of the joint decision objective function includes: constructing a first sub-objective of maximizing grid ancillary service revenue, calculating the total service fees obtained in response to grid dispatch within a future time window, calculated as the sum of the products of the dispatch decision variables for each time period, the provided power, and the grid price; constructing a second sub-objective of maximizing the battery's life-cycle value, obtaining the battery health value by calculating the product of the capacity degradation reduced by capacity restoration operations and the battery replacement cost, calculated as the sum of the products of the decision variables for performing capacity restoration in each time period, the reduced capacity degradation in that time period, and the battery replacement cost; and normalizing the two sub-objectives using... The mean normalization method based on the range maps the two sub-objectives to the same numerical interval. The two normalized sub-objectives are then weighted and fused. The joint decision objective function is the product of the first weight coefficient and the normalized grid ancillary service revenue, plus the product of the second weight coefficient and the normalized battery life cycle value. The sum of the first weight coefficient and the second weight coefficient is one. The value range of the first weight coefficient is 0.3 to 0.8, and the value range of the corresponding second weight coefficient is 0.2 to 0.7. When focusing on economic benefits, the value of the first weight coefficient is greater than 0.6, and when focusing on battery health, the value of the second weight coefficient is greater than 0.6.

[0014] Further, the method for calculating the capacity decay is as follows: based on the aging feature prediction sequence, a battery capacity prediction trajectory is generated when capacity recovery is not performed; a battery capacity prediction trajectory is generated after capacity recovery is performed during the time period; the capacity decay reduced by the capacity recovery operation is the capacity difference between the two trajectories at the end of the evaluation time window; the method of using dynamic programming to solve the bi-objective optimization problem, and solving the optimal decision of each stage in reverse from the end of the time window, includes: dividing the future time window into multiple decision stages, and solving the optimal decision of each stage in reverse from the last stage; in each scenario, starting from the end of the time window, calculating the objective function value of various decision choices under the premise of satisfying the recovery hard constraint, the decision choices include responding to the power grid. The system schedules or performs capacity restoration. If capacity restoration is performed, the temperature parameter is selected from the robust temperature parameter candidate set. The optimal cumulative benefit from the time period to the end of the current state is recorded through a value function. The state includes battery capacity and whether the hard constraint of restoration is met. The recursion of the value function is that the optimal value of the current time period is equal to the maximum value of the sum of the immediate benefits under each decision action and the value function of the next time period. When the decision action is to respond to grid dispatch, the immediate benefit is the product of the first weight coefficient and the dispatch decision variable, the power provided, and the grid price. When the decision action is to perform capacity restoration, the immediate benefit is the product of the second weight coefficient and the capacity restoration decision variable, the capacity degradation amount, and the battery replacement cost. The system recursively calculates back to the current time to obtain the complete optimal decision sequence for this scenario.

[0015] Further, the step of extracting the execution instructions for the current time period from the robust optimal decision scheme includes: extracting the execution instructions for the current time period from the robust optimal decision scheme, converting the decision variables into executable control instructions; for time period scheduling decision variables, when the decision variable takes the value of one, converting it into an instruction to perform capacity recovery operation, and when the decision variable takes the value of zero, converting it into an instruction to respond to grid scheduling; for temperature parameter decision variables, converting the temperature values ​​in the robust temperature parameter candidate set corresponding to the decision variable into the target temperature setpoint of the thermal management system; the execution instructions include the scheduling decision for the current time period and the corresponding control parameters; if capacity recovery is performed, the control parameters include robust temperature control parameters and thermal management control margin; the thermal management control margin is determined based on the uncertainty range of the ambient temperature and the power capacity of the battery thermal management system, and is calculated as the smaller value of the uncertainty range of the ambient temperature and the power capacity of the battery thermal management system divided by the product of the battery specific heat capacity and the battery mass.

[0016] This invention explicitly models the multidimensional uncertainties of grid ancillary service demand and ambient temperature using a scenario tree generation algorithm, assigning probability weights to each possible scenario. This overcomes the problem of traditional deterministic optimization methods treating predicted values ​​as single deterministic values, leading to sensitivity to prediction errors. By performing sensitivity analysis on the battery temperature-capacity recovery efficiency function curve, it identifies low-sensitivity regions where capacity recovery efficiency is insensitive to temperature fluctuations as a robust temperature parameter candidate set. This overcomes the problem of traditional point-optimized temperature methods, which rely on constant-temperature environment calibration, resulting in a significant decrease in capacity recovery efficiency under environmental disturbances. Furthermore, by using dynamic programming to solve a bi-objective optimization problem constrained by robust temperature in each scenario, it locates periods of low grid demand to schedule capacity recovery operations while satisfying the hard constraint of battery health, thus coordinating grid service response with capacity recovery timing. Finally, through robust optimization criteria, it selects the worst-case objective value from the set of optimal decision schemes, ensuring that the decision scheme maintains acceptable economic benefits and battery health levels even under uncertainty propagation. This invention addresses the technical problems of capacity recovery decision-making failing to consider the coordinated optimization of grid ancillary service demand and battery health, and the high vulnerability of decision-making schemes in uncertain environments due to optimization based on deterministic predictions. It achieves the joint optimization of capacity recovery timing, temperature parameters, and grid service response under the dual uncertainties of grid demand and environmental conditions, thereby improving the robustness and reliability of decision-making schemes in balancing economic benefits and battery health in uncertain environments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a capacity recovery method based on the entire life cycle of a lithium-ion battery according to the present invention. Detailed Implementation

[0018] This embodiment provides a capacity recovery method based on the entire life cycle of lithium-ion batteries. This capacity recovery method is applied to the battery management system of an energy storage power station. The energy storage power station includes a battery array, a battery thermal management system, and an energy storage system controller. The battery management system is connected to the power grid dispatching system and the weather forecasting system for data transmission.

[0019] like Figure 1 As shown, it includes the following steps: Step 100: Obtain the forecast data within the future time window and generate the uncertainty parameter set. Obtain forecast data for grid ancillary service demand and ambient temperature within a future time window. The grid ancillary service demand forecast data includes time series data for peak-shaving period distribution, frequency regulation demand intensity, and ancillary service prices. The ambient temperature forecast data includes predicted ambient temperature values ​​for each future time period. Simultaneously, obtain the uncertainty range for each forecast variable: the uncertainty range for grid prices is the price fluctuation range, and the uncertainty range for ambient temperature is the temperature forecast error range.

[0020] Furthermore, the duration of the future time window is determined based on the power grid dispatching plan cycle, and its value range is [value range missing]. Hours to Hour.

[0021] The power grid ancillary service demand forecast data, ambient temperature forecast data, and the corresponding uncertainty range are organized into an uncertainty parameter set.

[0022] It should be noted that grid ancillary service demand forecast data can be obtained from historical dispatch records and future demand plans of the grid dispatching system. The fluctuation range of grid prices can be obtained by analyzing the fluctuation amplitude of historical data for the same period. Environmental temperature forecast data can be obtained from the meteorological forecasting system, providing hourly temperature forecasts for the next week. The temperature forecast error range is determined based on the forecast confidence interval provided by the meteorological forecasting system. For example, if the forecast temperature is... At that time, the uncertainty range was set to ,in This represents the statistical value of historical forecast errors.

[0023] In this embodiment of the application, to quantify uncertainty more precisely, in addition to obtaining the range of the predictor variables, the probability distribution characteristics of each predictor variable within the uncertainty range are also obtained. For example, for environmental temperature prediction, not only the temperature range is obtained... It also obtains the probability density function parameters that the temperature follows within this range, generating a probabilistic set of uncertainty parameters.

[0024] certain The battery management system of the MWh energy storage power station is moon day Get the future at any time Hourly forecast data. Ancillary service price forecast sequences are obtained from the power grid dispatching system, and ambient temperature forecast sequences are obtained from the weather forecasting system. Based on statistical analysis of historical fluctuations, the fluctuation range of power grid prices is determined as the forecast value. The prediction error range for ambient temperature is... ℃. Selecting the future Table 1 shows data for some time periods within an hour.

[0025] Table 1. Forecast data and range of uncertainty for future periods

[0026] The acquired power grid price forecast sequence, price fluctuation range, ambient temperature forecast sequence, and temperature error range are organized into an uncertainty parameter set, which serves as the input data for subsequent scenario tree generation.

[0027] Step 200: Acquire battery excitation response data and generate a robust temperature parameter candidate set. The excitation response data of the battery at multiple temperature points was acquired. The battery temperature was stabilized at different temperature points by controlling the battery thermal management system. At each temperature point, a capacity recovery operation was applied to the battery, and the capacity recovery efficiency was measured. Using a function fitting method, discrete temperature points and corresponding capacity recovery efficiency data were used as input to fit a continuous function curve of capacity recovery efficiency versus temperature. As output, Indicates capacity recovery efficiency. This indicates the battery temperature during the recovery operation.

[0028] Furthermore, the function fitting method employs polynomial least squares fitting, and the fitting function has the following form: ,in For the summation variable, For polynomial coefficients, Let be the order of the polynomial, and its range is . to The polynomial coefficients are determined by minimizing the sum of squared fitting errors.

[0029] Continuous function curve of recovery efficiency versus temperature Sensitivity analysis was performed to calculate the gradient of capacity recovery efficiency with respect to temperature deviation at each temperature point. Identify temperature ranges where the absolute value of the gradient is less than the sensitivity threshold. These temperature ranges are insensitive to temperature fluctuations in terms of volume recovery efficiency and are marked as low-sensitivity regions.

[0030] Furthermore, the gradient By fitting the function We obtain this by analytical differentiation for polynomial functions. Its derivative is ,in For the summation variable.

[0031] Furthermore, the sensitivity threshold is determined based on the temperature control accuracy of the battery thermal management system, and the calculation formula is as follows: ,in The maximum allowable rate of change in capacity recovery efficiency, with a value range of [value missing]. to , This refers to the temperature control error of the battery thermal management system.

[0032] In the low-sensitivity region, the average comprehensive computing capacity recovery efficiency and standard deviation of capacity recovery efficiency Z-score standardization was applied to the mean and standard deviation of capacity recovery efficiency to eliminate the influence of dimensions, and robustness evaluation index was used. An assessment was conducted, in which and These represent the standardized mean and standard deviation of capacity recovery efficiency, respectively. This is the risk aversion coefficient. The range of values ​​is to Select temperature points where the robustness evaluation index value is higher than the robustness threshold to generate a candidate set of robust temperature parameters.

[0033] Furthermore, the robustness threshold is set based on the minimum capacity recovery efficiency requirement for capacity recovery, and the formula for calculating the robustness threshold is as follows: ,in The minimum permissible capacity recovery efficiency after standardization. The standard deviation of the maximum permissible capacity recovery efficiency after standardization. The value of is determined based on the maximum value of the standard deviation of capacity recovery efficiency in the low-sensitivity region.

[0034] It should be noted that the above sensitivity thresholds are determined based on the temperature control accuracy of the battery thermal management system. For example, the temperature control error of the battery thermal management system is... ℃, then the sensitivity threshold is set to allow the temperature deviation to be within acceptable limits. The change in capacity recovery efficiency at ℃ does not exceed The gradient value. The robustness threshold is set based on the minimum capacity recovery efficiency requirement for capacity recovery, ensuring that the temperature parameters in the candidate set can still meet basic recovery requirements even under uncertain environments.

[0035] In this embodiment, to adapt to the changes in battery temperature response characteristics at different aging stages, the current aging state parameters of the battery are also acquired when acquiring the excitation response data. The temperature-capacity recovery efficiency function curve is piecewise fitted based on the aging state parameters, generating a corresponding robust temperature parameter candidate set for different aging stages to improve the applicability of the robust parameters.

[0036] The battery management system of this energy storage power station controls the battery temperature through the battery thermal management system. ℃ to Every ℃ range A temperature test point was set at ℃, and the capacity recovery efficiency was measured at each temperature point. The test data are shown in Table 2.

[0037] Table 2. Capacity recovery efficiency test data at different temperature points

[0038] The fitting function is obtained by using third-order polynomial least squares fitting. The gradient function is obtained by differentiating the fitted function. The temperature control error of the battery thermal management system is... ℃, set the maximum allowable rate of change in capacity recovery efficiency The sensitivity threshold is calculated as follows: ℃. Calculate the gradient value at each temperature point and identify... ℃ to ℃ represents a low-sensitivity region. Within this region, the average capacity recovery efficiency is... Standard deviation Set a risk aversion coefficient. Calculate robustness evaluation index (Normalized value) is above the robustness threshold. Temperature points in low-sensitivity regions ℃ ℃ ℃ ℃ added to the robust temperature parameter candidate set ℃.

[0039] Step 300: Obtain the battery aging feature prediction sequence and generate recovery time flexibility labels. Obtain the battery aging characteristic prediction sequence, which includes capacity decay rate, internal resistance growth rate, and polarization voltage change rate.

[0040] Furthermore, the aging characteristic prediction sequence is obtained through historical monitoring data from the battery management system. Based on the battery's historical capacity, internal resistance, and polarization voltage data, a time series prediction model is used to generate predicted aging characteristic values ​​within future time windows. The capacity decay rate is calculated as the ratio of the current capacity to the initial capacity, the internal resistance growth rate is calculated as the ratio of the current internal resistance to the initial internal resistance, and the polarization voltage change rate is calculated as the ratio of the difference in polarization voltage between adjacent time periods to the time interval.

[0041] Based on aging characteristic prediction sequences, the time points when the capacity decay rate exceeds the decay threshold and the time points when the internal resistance growth rate exceeds the internal resistance threshold are analyzed to identify the mandatory execution time window for capacity recovery operations. The mandatory execution time window is a hard constraint, indicating that the capacity recovery operation must be completed within this time window; otherwise, irreversible degradation of battery performance will occur.

[0042] Furthermore, the attenuation threshold is the percentage decrease in battery capacity relative to the initial capacity, and the range of the attenuation threshold is [range missing]. to The internal resistance growth rate threshold is the percentage increase in battery internal resistance relative to the initial internal resistance, and the value range of the internal resistance growth rate threshold is [value missing]. to .

[0043] Simultaneously, the analysis identifies the optimal execution time window for capacity recovery by analyzing the time points when the capacity decay rate approaches the decay threshold. The optimal execution time window is a soft constraint, meaning that performing capacity recovery operations within this window will achieve good recovery results, but delaying execution will not lead to irreversible damage. Flexibility weights are assigned to the mandatory execution time windows. Assign flexibility weights to the optimal execution time window. Generate recovery time flexibility tags.

[0044] Furthermore, the flexibility weight The range of values ​​is to When the difference between the capacity decay rate and the decay threshold is smaller, The larger the value, the better.

[0045] It should be noted that the degradation threshold and internal resistance threshold are determined based on battery specifications and application scenarios. For example, for energy storage applications, the capacity degradation rate exceeding the initial capacity is considered a threshold. The time is set as a hard constraint that requires recovery to be performed. Flexibility weights. The range of values ​​is to The smaller the value, the higher the scheduling flexibility of the preferred execution time window.

[0046] The initial battery capacity of this energy storage power station is MWh, currently running Months. The battery management system generates future forecasts using a time series prediction model based on historical monitoring data. Aging characteristic prediction sequence for hours. Current battery capacity is... MWh, capacity decay rate The current internal resistance is mΩ, internal resistance growth rate is Set the attenuation threshold to... The internal resistance growth rate threshold is The predictive model shows that if no capacity recovery operation is performed, the capacity degradation rate will be... moon day achieve Exceeding the decay threshold; the internal resistance growth rate will be moon day achieve This exceeds the internal resistance threshold. Therefore, the necessary execution time window is identified as [missing information]. moon day Previously, flexibility weights were assigned. Simultaneously, analysis revealed that the capacity decay rate was... moon day achieve Approaching the decay threshold, the optimal execution time window is identified as... moon day to moon day The flexibility weight is calculated based on the difference between the capacity decay rate and the threshold. Generate recovery time flexibility labels, which serve as constraint inputs for subsequent optimization decisions.

[0047] Step 400: Generate the scene tree structure using a scene tree generation algorithm. A scene tree generation algorithm is used to generate multiple possible future scenes based on a set of uncertain parameters. The input of the scene tree generation algorithm is the set of uncertain parameters, and the output is a scene tree structure.

[0048] A Monte Carlo sampling method is used to generate a set of scenarios. The Monte Carlo sampling method takes the probability distribution of each uncertain parameter as input and outputs multiple scenario samples. Before sampling, the ambient temperature and grid price are normalized based on their range to map parameters with different dimensions and numerical ranges to the same numerical interval, eliminating the influence of dimensional differences on sampling. Random sampling is performed based on the normalized probability distribution of the ambient temperature and grid price, generating one scenario per sampling. Each scenario corresponds to a specific combination of uncertain parameters, including the ambient temperature sequence and grid price sequence for that scenario. Scenario types include high temperature and high electricity price scenarios, low temperature and low electricity price scenarios, high temperature and low electricity price scenarios, and low temperature and high electricity price scenarios.

[0049] A scene reduction algorithm is used to filter the scene set. The input of the scene reduction algorithm is the scene set generated by Monte Carlo sampling, and the output is a representative scene set. The similarity between scenes is calculated in the normalized parameter space. Representative scenes with large probability weights and significant differences are retained. Similar scenes are merged and their probability weights are accumulated to reduce the number of scenes.

[0050] Furthermore, the similarity between the scenes is calculated using Euclidean distance. and scene The formula for calculating the similarity between them is: ,in and For scene indexing, For time period, and Scenes During the period The normalized temperature and normalized price. When Scenes with similarity values ​​less than a similarity threshold are considered similar. The similarity threshold ranges from the diagonal length of the normalized parameter space. to The criterion for determining a larger probability weight is that the probability weight of a scene is higher than the average probability weight of all scenes.

[0051] After the screening is completed, the parameters of the representative scenarios are denormalized to restore the original temperature and price values.

[0052] Based on the probability distribution of each parameter, probability weights are assigned to each representative scenario. Probability weights Representing a scene The probability of occurrence, the sum of the probability weights of all scenarios is .

[0053] Furthermore, the scene probability weights are determined by the frequency of the scene's occurrence in the Monte Carlo sampling, for representative scenes... Its probability weights ,in To be merged into the scene The number of original scenes, The total number of scenes generated for Monte Carlo sampling.

[0054] The scenarios and their probability weights are organized into a tree structure. The root node of the tree represents the current moment, the branch nodes represent future time periods, and each branch corresponds to a scenario evolution path, thus generating a scenario tree structure.

[0055] In this embodiment, to capture the correlation between uncertain parameters, the historical correlation between ambient temperature and grid demand is analyzed when generating the scene. For example, high temperatures are often accompanied by peak electricity consumption, leading to higher grid prices. This correlation is incorporated into the joint probability distribution of scene generation, making the generated scene more consistent with reality.

[0056] Based on the set of uncertain parameters, the Monte Carlo method is used to generate A sample of scenarios. Before sampling, the ambient temperature and power grid price were normalized to define the temperature range. ℃ mapped to Price range Yuan / MWh mapping to Random sampling is performed based on the probability distribution of each parameter to generate a scene set containing ambient temperature and power grid price sequences. A scene reduction algorithm is used to calculate the Euclidean distance between scenes in the normalized parameter space, with a similarity threshold set as follows. (Diagonal length of the normalized parameter space) Similar scenes are merged. Those selected are retained. For each representative scenario, the parameters are denormalized to restore their original dimensions. The characteristic parameters of the representative scenarios are shown in Table 3.

[0057] Table 3. Representative Scene Feature Parameters

[0058] Will The representative scenarios and their probability weights are organized into a tree structure, with the root node representing the current time step. moon day Branch nodes represent the future. Within each hour, each branch corresponds to the evolution path of a scene, generating a scene tree structure as input for subsequent optimization.

[0059] Step 500: Construct a joint decision objective function to generate an economic-health dual-objective optimization problem. A joint decision-making objective function is constructed, comprising two sub-objectives. The first sub-objective is to maximize the revenue from grid ancillary services, calculating the total service costs obtained in response to grid dispatch within a future time window. The second sub-objective is to maximize the battery's total lifecycle value, considering the contribution of capacity recovery to battery life extension. This is achieved by calculating the battery health value by multiplying the reduced capacity degradation from capacity recovery operations by the battery replacement cost.

[0060] The two sub-objectives are normalized using a mean normalization method based on the range. and By mapping each sub-objective to the same numerical interval, the influence of the difference in the magnitude of the two sub-objectives on the weighted fusion is eliminated. The two normalized sub-objectives are then weighted and fused, and the joint decision objective function is expressed as follows: ,in and These are the normalized grid ancillary service revenue and the battery's total lifecycle value, respectively. and The weighting coefficients and .

[0061] Furthermore, the weighting coefficients The range of values ​​is to Correspondingly The range of values ​​is to When focusing on economic benefits, Value greater than When focusing on battery health, Value greater than .

[0062] The decision variables include whether to respond to grid dispatch or perform capacity restoration in each time period, and the temperature parameters used during capacity restoration. Constraints include the hard constraint that restoration must be satisfied and the temperature parameters must be within the robust temperature parameter candidate set, generating an economic-health bi-objective optimization problem.

[0063] Furthermore, the specific form of the constraint conditions includes: the first constraint condition is a time-period mutual exclusion constraint, , indicating time period It cannot simultaneously respond to grid dispatch and perform capacity restoration; the second constraint is a hard constraint for restoration. ,in The time period indicates the time window during which execution is required. At least one capacity recovery operation must have been performed previously; the third constraint is a temperature parameter constraint. This indicates the temperature parameter during capacity recovery. Must be in the robust temperature parameter candidate set The fourth constraint is the constraint on the values ​​of the decision variables. , indicating that the decision variable is a binary variable.

[0064] It should be noted that the weighting coefficients and The decision is determined based on the energy storage power station's operational strategy. For operational models that prioritize economic returns, the following settings are established. Larger, for example For operating models that prioritize battery health, set... Larger, for example Revenue from power grid ancillary services The calculation formula is ,in For time period, and These represent the start and end times of the future time window, respectively. For time period Scheduling decision variables, For the power provided, For grid price. Battery lifecycle value. The calculation formula is ,in For time period, For time period Decision variables for implementing capacity recovery For capacity recovery operations during the time period Reduced capacity decay Battery replacement costs.

[0065] Furthermore, the duration of the future time window is determined based on the power grid dispatch cycle and the battery aging rate, and the value range of the time window duration is as follows: Hours to The time window is divided into multiple time periods, with each time period lasting for hours. minutes to Hours. The power provided. The value is determined based on the rated power of the energy storage power station and the grid demand, and the range is [amount missing] of the rated power of the energy storage power station. to The battery replacement cost Calculated based on battery purchase price, installation cost, and residual value. ,in For battery purchase price, For installation costs, To recover residual value.

[0066] Furthermore, the capacity attenuation amount The calculation method is as follows: Based on the aging feature prediction sequence, a battery capacity prediction trajectory is generated when capacity recovery is not performed. Generated in time period Battery capacity prediction trajectory after capacity recovery The capacity reduction caused by the capacity recovery operation is the capacity difference between the two trajectories at the end of the evaluation time window, calculated using the following formula: ,in To evaluate the end of the time window.

[0067] Step 600: Solve the optimization problem using dynamic programming algorithm to generate a set of conditionally optimal decision solutions. In each branch of the scene tree structure, a dynamic programming algorithm is used to solve the bi-objective optimization problem. The input of the dynamic programming algorithm includes the objective function of the bi-objective optimization problem, decision variables, constraints, temperature and price sequences of each scene in the scene tree structure, robust temperature parameter candidate set, and recovery time flexibility label. The output is the conditionally optimal decision sequence for each scene.

[0068] The dynamic programming algorithm divides the future time window into multiple decision-making stages and works backward from the last stage to find the optimal decision for each stage. In each scenario, starting from the end of the time window, the objective function value of various decision choices is calculated, while satisfying the recovery hard constraint. Decision choices include responding to grid dispatch or performing capacity restoration; if capacity restoration is performed, the temperature parameter must be selected from a robust set of temperature parameter candidates. By comparing the objective function values ​​of different decision choices, the decision that maximizes the objective function is selected as the optimal decision for that scenario and that time period. Working backward to the current time, the complete optimal decision sequence for that scenario is obtained.

[0069] Repeat the above process for all branches of the scenario tree structure to generate a set of conditionally optimal decision schemes for each scenario, where each scheme corresponds to an optimal decision sequence for a scenario.

[0070] It's important to note that during the state transition process in dynamic programming, the decisions made in the current time period affect the feasible decision space in subsequent time periods. For example, if the failure to perform capacity restoration in the current time period leads to a capacity decay rate exceeding a threshold, the hard constraint that capacity restoration must be performed in subsequent time periods is activated. The dynamic programming algorithm utilizes value functions... Record from time period To the end in state The optimal cumulative return under the given state This includes information such as battery capacity and whether the recovery hard constraint is met. The recursive formula for the value function is... ,in For time period Decision-making actions For time period In state Make an executive decision The immediate benefits gained To implement the decision The state then transitions to the next state. The boundary condition is that the value function of the last time period equals the immediate gain of that time period.

[0071] Furthermore, the immediate benefits The calculation method is as follows: when the decision action is to respond to grid dispatch, the immediate benefit is... When the decision action is to restore capacity, the immediate benefit is... ;in and These are the normalized weighting coefficients.

[0072] In this embodiment of the application, in order to improve the solution efficiency, a state space discretization method is adopted in the dynamic programming process to discretize the continuous battery capacity state into a finite number of state points, and an approximate dynamic programming algorithm is used to process the high-dimensional state space to reduce computational complexity.

[0073] The rated power of the energy storage power station is MW, the output power range when providing ancillary services is MW to MW. Battery procurement price Ten thousand yuan, installation cost 10,000 yuan, residual value recovered The battery replacement cost is calculated to be 10,000 yuan. The energy storage power station adopts an operating strategy that balances economic benefits and battery health, setting a weighting coefficient. , According to the recovery time flexibility label in step 300, it must be... moon day Perform capacity recovery operations beforehand. (This will be done in the future.) Hours are divided into There are 1 time periods, each with a duration of 1 hour. Minutes. In scenario S1, a dynamic programming algorithm is used to solve the bi-objective optimization problem. Starting from the final time period... moon day We begin the reverse recursive process, calculating the optimal decision for each time period under the hard constraints. Temperature parameters must be selected from a robust set of temperature parameter candidates. The decision was selected based on the temperature range (℃). The decision results for some time periods are shown in Table 4.

[0074] Table 4 Optimization decision results for a portion of the time period in scenario S1

[0075] In scenario S1, the optimized algorithm identifies moon day For periods with relatively low prices and before the preferred execution time window, capacity recovery operations are performed during this period, and temperature parameters are selected from a robust temperature parameter candidate set. ℃. Performing capacity recovery during this period can reduce capacity decay. MWh, battery health value is The immediate return after normalization is 10,000 yuan. Ten thousand yuan. For all elements in the scene tree structure... Repeat the dynamic programming solution process for each scenario to generate a set of conditionally optimal decision solutions for each scenario.

[0076] Step 700: Select the robust optimal decision scheme using the robust optimization criterion. For each solution in the set of optimal decision solutions, calculate the performance of each solution under all scenarios. The solution is applied to each scene in the scene tree structure, and the objective function value of the solution is calculated in each scene. Calculate the weighted average target value. ,in For the scene, For the scene The probability weights. Calculate the worst-case target value. .

[0077] The robust optimization criterion is used to select the robust optimal decision scheme. The robust optimization criterion is to minimize the weighted sum of economic and health losses in the worst-case scenario, expressed as selecting the scheme that minimizes the economic and health losses. The optimal solution ensures that acceptable target values ​​can be achieved even in the most unfavorable scenarios.

[0078] Furthermore, the selection process for the robust optimal decision scheme is as follows: traverse all schemes in the set of conditionally optimal decision schemes, calculate the worst-case objective value for each scheme under all scenarios, and select the scheme with the largest worst-case objective value as the robust optimal decision scheme. The calculation formula is as follows: ,in The selected robust optimal decision scheme.

[0079] Generate robust optimal decision schemes.

[0080] It should be noted that robust optimization criteria can be adjusted according to the decision-maker's risk appetite. Besides the conservative criterion of minimizing the worst-case loss, a conditional risk value criterion can also be used, considering the average performance of adverse scenarios with higher probability weights. The robust optimization criterion is expressed as follows: ,in This is the risk preference coefficient. The larger the value, the more importance is placed on average performance. The smaller the value, the stronger the need for defense against the worst-case scenario.

[0081] Robustness evaluation is performed on the decision schemes for each scenario in the set of optimal decision schemes. The optimal decision schemes for each scenario are then applied to all scenarios. In each scenario, the objective function values ​​of each solution are calculated under different scenarios. The performance evaluation of each solution is shown in Table 5.

[0082] Table 5 Objective function values ​​of each decision scheme in different scenarios

[0083] According to the robust optimization criterion, the solution with the largest worst-case objective value is selected. As shown in Table 5, the worst-case objective value of solution S2 in all scenarios is... The target value is 10,000 yuan, higher than the worst-case value of other options. Although the weighted average target value of option S2 is... While not the highest price, it offers the most stable performance in unfavorable scenarios, ensuring consistent results even in the most challenging S2 conditions. The acceptable return is 10,000 yuan. Therefore, option S2 is chosen as the robust optimal decision-making option, which is arranged in... moon day Perform capacity recovery operation, select temperature parameters. ℃.

[0084] Step 800: Extract the execution command and output it to the control system. Extracting execution instructions for the current time period from a robust optimal decision scheme. The decision variables in the robust optimal decision scheme include time period scheduling decision variables and temperature parameter decision variables. These decision variables are then converted into executable control instructions. For the time period scheduling decision variables, when the decision variable takes the value... When the decision variable is set to a value, it is converted into an instruction to perform a capacity restoration operation. The time frame is converted into instructions to respond to grid dispatch. For temperature parameter decision variables, the temperature values ​​in the robust temperature parameter candidate set corresponding to the decision variable are converted into the target temperature setpoint of the thermal management system. The execution instructions include the dispatch decision for the current time period and the corresponding control parameters. The dispatch decision indicates whether to respond to grid dispatch or perform capacity restoration for the current time period. If capacity restoration is performed, the control parameters include robust temperature control parameters and thermal management control margin. The robust temperature control parameters are the target temperature values ​​selected from the robust temperature parameter candidate set, and the thermal management control margin is the temperature regulation margin reserved by the thermal management system to cope with environmental disturbances.

[0085] Furthermore, the thermal management control margin is determined based on the uncertainty range of the ambient temperature and the power capacity of the battery thermal management system, and the calculation formula is as follows: ,in Given the range of uncertainty in ambient temperature, For the power capacity of the battery thermal management system, For the specific heat capacity of the battery, For battery quality.

[0086] The execution instructions are output to the energy storage system controller and the battery thermal management system. The energy storage system controller executes grid dispatch response or initiates capacity restoration procedures based on dispatch decisions. The battery thermal management system adjusts the battery temperature based on robust temperature control parameters and thermal management control margin.

[0087] It should be noted that the thermal management control margin is determined based on the uncertainty range of the ambient temperature. For example, the uncertainty range of the ambient temperature is... The target temperature set by the battery thermal management system is ℃. The actual temperature control range is then set to ,in This is the control margin calculated based on the power capacity of the battery thermal management system and the magnitude of environmental disturbances.

[0088] In this embodiment, to improve the operability of the execution instructions, alternative execution instructions are generated during the extraction of the execution instructions. When the main execution instruction cannot be executed due to actual limitations, it automatically switches to the alternative instructions, thereby improving the adaptability of the decision-making scheme.

[0089] Extracting the current time period from the robust optimal decision solution moon day to The execution command. According to the S2 scheme, the current time period's scheduling decision is "respond to grid dispatch." After receiving the command, the energy storage system controller prepares to respond to the grid's frequency regulation requirements, and the output power is set to... MW. Robust optimal decision-making schemes show that, in moon day A capacity recovery operation needs to be performed during the specified time period. The robust temperature control parameters are as follows: ℃. The uncertainty range of ambient temperature is [missing information]. ℃, battery thermal management system power capacity kW, battery specific heat capacity J / (kg·℃), total battery mass kg, calculate thermal management control margin ℃. The battery thermal management system is set to a temperature control range of ℃. ℃, to ensure that the ambient temperature fluctuates It can maintain the battery temperature at the target temperature even at ℃. The temperature is around ℃. The complete execution instructions for the current time period are shown in Table 6.

[0090] Table 6 Execution Instructions for the Current Time Period

[0091] The execution command is output to the energy storage system controller and the battery thermal management system. The energy storage system controller executes grid frequency regulation response based on dispatch decisions, providing... MW's ancillary services. The battery thermal management system receives temperature parameters and control margin information to prepare for future capacity recovery operations by adjusting the temperature accordingly.

[0092] Step 900: Monitor actual values ​​in real time and trigger scrolling optimization. Real-time monitoring of actual grid demand, battery status, and ambient temperature. Actual grid demand includes actual dispatch commands and actual service prices; actual battery status includes actual capacity and internal resistance; and actual ambient temperature is the ambient temperature measured by the battery thermal management system.

[0093] Calculate the deviation between the actual and predicted values. When the deviation between the actual and predicted grid price exceeds the price deviation threshold, or the deviation between the actual and predicted ambient temperature exceeds the temperature deviation threshold, the prediction deviation is determined to be significant.

[0094] Furthermore, the price deviation threshold is determined based on the fluctuation range of the power grid price within the uncertainty parameter set, and the calculation formula is as follows: ,in and These are the upper and lower bounds for predicting grid prices, respectively. This is the price deviation coefficient, with a value range of [value range missing]. to The temperature deviation threshold is determined based on the prediction error range of the ambient temperature in the uncertainty parameter set, and the calculation formula is as follows: ,in This represents the statistical value of temperature prediction error. This is the temperature deviation coefficient, with a value range of [value range missing]. to .

[0095] Update the uncertainty range of the uncertainty parameter set, expand the uncertainty range according to the actual deviation, and update the probability weight of each scenario in the scenario tree structure. Reduce the weight of scenarios that deviate significantly from the actual situation and increase the weight of scenarios that are close to the actual situation.

[0096] Furthermore, the method for updating the scene probability weights is as follows: calculate the deviation between the parameter values ​​and the actual values ​​for each scene. The deviation index is ,in and These are the actual temperature and the actual price, respectively. and For the scene Temperature and price. Based on the deviation index, the probability weights are updated using an exponential weighting method. The updated probability weights are: ,in Represents an exponential function. For the scene, This is the weighting adjustment factor, and its value range is... to .

[0097] Trigger rolling optimization, re-execute steps 400 to 700, recalculate the scheduling scheme for subsequent time periods based on the updated scene tree structure, and generate an updated robust optimal decision scheme.

[0098] It should be noted that the price deviation threshold and temperature deviation threshold are determined based on the initial setting range of the uncertainty parameter set. For example, the price deviation threshold is set to the width of the initial price fluctuation range. When the actual deviation exceeds this threshold, it indicates that the prediction uncertainty has been underestimated and re-optimization is required. The trigger frequency of rolling optimization is determined based on the duration of the actual deviation, avoiding the waste of computational resources caused by frequent optimization triggers due to short-term fluctuations.

[0099] During execution, the battery management system monitors in real time. moon day Actual data for the time period. The power grid dispatching system displays the actual ancillary service price as follows: Yuan / MWh, the actual ambient temperature measured by the ambient temperature monitoring system is ℃. Calculate the deviation between the actual value and the predicted value; the predicted grid price is... Yuan / MWh, with a deviation of Yuan / MWh. Based on the uncertainty parameter set, the width of the grid price fluctuation range is... Yuan / MWh, price deviation coefficient Price deviation threshold Yuan / MWh. Actual price deviation. Yuan / MWh did not exceed the threshold Yuan / MWh. The predicted ambient temperature is [value missing]. ℃, actual deviation is ℃, temperature deviation coefficient Temperature deviation threshold The temperature deviation is within the threshold. (℃) moon day During the time period, the actual price for ancillary services is Yuan / MWh, predicted value Yuan / MWh, with a deviation of The price per MWh is significantly higher than the price deviation threshold, indicating a significant prediction error and triggering the rolling optimization mechanism. The deviation index between each scenario and the actual situation is calculated; the average price for scenario S1 is... The average price of S2 in scenario S2 is [price per MWh]. Yuan / MWh, actual price The per-yuan / MWh ratio more closely resembles the characteristics of scenario S1 under high electricity prices, and scenario S1 exhibits the smallest deviation. A weighting adjustment coefficient is then set. The probability weights of scenario S1 are updated using an exponential weighting method. Upgraded to The probability weights of scenario S2 are from Reduce to At the same time, the uncertainty range of power grid prices is expanded based on actual deviations, with the fluctuation range extending from... Adjusted to Based on the updated scene tree structure, steps 400 to 700 are re-executed to generate an updated robust optimal decision scheme.

Claims

1. A capacity recovery method based on the entire life cycle of a lithium-ion battery, characterized in that, Includes the following steps: Obtain forecast data on power grid ancillary service demand and ambient temperature within a future time window, along with the corresponding uncertainty range, and organize them into a set of uncertainty parameters. Capacity recovery efficiency data of the battery at multiple temperature points is obtained, and a continuous function curve of capacity recovery efficiency versus temperature is fitted. Sensitivity analysis is performed on the function curve, and the gradient of capacity recovery efficiency with temperature deviation at each temperature point is calculated. Temperature ranges with absolute gradient values ​​less than the sensitivity threshold are identified as low-sensitivity regions. In the low-sensitivity regions, robustness evaluation indexes are used for evaluation. Temperature points with robustness evaluation index values ​​higher than the robustness threshold are selected to generate a robust temperature parameter candidate set. Obtain the battery aging feature prediction sequence, identify the necessary execution time window and the preferred execution time window for capacity recovery operation based on the aging feature prediction sequence, assign a first flexibility weight to the necessary execution time window, assign a second flexibility weight to the preferred execution time window, and generate a recovery time flexibility label. Using a scene tree generation algorithm, a scene set is generated based on an uncertainty parameter set and a Monte Carlo sampling method. A scene reduction algorithm is used to filter the scene set to obtain a representative scene set. A probability weight is assigned to each representative scene, and the scenes and their probability weights are organized into a tree structure to generate a scene tree structure. A joint decision objective function is constructed, which includes two sub-objectives: maximizing the revenue of grid ancillary services and maximizing the value of the battery throughout its life cycle. The two sub-objectives are weighted and fused. The decision variables include the scheduling decisions for each time period and the temperature parameters used during capacity recovery. The constraints include hard constraints on recovery and the requirement that the temperature parameters must be within the robust temperature parameter candidate set, thus generating an economic-health bi-objective optimization problem. In each branch scenario of the scenario tree structure, the dynamic programming algorithm is used to solve the bi-objective optimization problem. Starting from the end of the time window, the optimal decision of each stage is solved in reverse to generate the set of conditionally optimal decision schemes for each scenario. For each scheme in the set of optimal decision schemes, calculate the worst-case objective value of each scheme under all scenarios, and select the scheme with the largest worst-case objective value as the robust optimal decision scheme. The execution instructions for the current period are extracted from the robust optimal decision scheme. The execution instructions include scheduling decisions and robust temperature control parameters. The execution instructions are then output to the energy storage system controller and the battery thermal management system.

2. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 1, characterized in that, Real-time monitoring of actual values ​​of grid demand, battery status, and ambient temperature; when the deviation between actual and predicted values ​​exceeds the deviation threshold, updating the probability weights of each scenario in the scenario tree structure, triggering rolling optimization, and recalculating the scheduling scheme for subsequent time periods. The robustness evaluation index is obtained by calculating the mean and standard deviation of capacity recovery efficiency in the low-sensitivity region, standardizing the mean and standard deviation of capacity recovery efficiency, and calculating the difference between the product of the standardized mean capacity recovery efficiency and the risk aversion coefficient and the standardized standard deviation of capacity recovery efficiency. The sensitivity threshold is determined based on the temperature control accuracy of the battery thermal management system and is calculated as the ratio of the maximum allowable rate of change of capacity recovery efficiency to the temperature control error.

3. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 2, characterized in that, Capacity recovery efficiency data of the battery at multiple temperature points were obtained, and a continuous function curve of capacity recovery efficiency versus temperature was fitted, including: By controlling the battery thermal management system, the battery temperature is stabilized at different temperature points, and a capacity recovery operation is applied to the battery at each temperature point, and the capacity recovery efficiency is measured. A polynomial least squares fitting method is employed, using discrete temperature points and corresponding capacity recovery efficiency data as input. The fitting function is a polynomial relationship between capacity recovery efficiency and temperature, with the polynomial order ranging from two to four. The polynomial coefficients are determined by minimizing the sum of squared fitting errors, resulting in a continuous function curve of capacity recovery efficiency versus temperature. Sensitivity analysis of the function curve was performed, and the gradient of capacity recovery efficiency with respect to temperature deviation at each temperature point was calculated, including: The continuous function curve of capacity recovery efficiency versus temperature obtained by fitting is analytically differentiated. For polynomial functions, the gradient function is obtained by differentiating each term. Calculate the gradient value at each temperature point, identify the temperature range where the absolute value of the gradient is less than the sensitivity threshold, and mark it as a low-sensitivity region. The sensitivity threshold is calculated by setting the maximum allowable rate of change in capacity recovery efficiency and dividing the maximum rate of change in capacity recovery efficiency by the temperature control error of the battery thermal management system. The maximum rate of change in capacity recovery efficiency ranges from 3% to 8%.

4. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 1, characterized in that, The evaluation using robustness metrics in the low-sensitivity region includes: Within the low-sensitivity region, calculate the mean and standard deviation of capacity recovery efficiency; Standardize the mean and standard deviation of capacity recovery efficiency to eliminate the influence of dimensions; The robustness evaluation index is used for assessment. The robustness evaluation index is calculated as the product of the standardized mean capacity recovery efficiency minus the risk aversion coefficient and the standard deviation of the standardized capacity recovery efficiency. The risk aversion coefficient ranges from 0.5 to 2. Temperature points with robustness evaluation index values ​​higher than robustness thresholds are selected to generate a candidate set of robust temperature parameters. The robustness threshold is calculated as the product of the standardized minimum allowable capacity recovery efficiency minus the risk aversion coefficient and the standardized maximum allowable capacity recovery efficiency standard deviation. The standardized maximum allowable capacity recovery efficiency standard deviation is determined based on the maximum value of the capacity recovery efficiency standard deviation in the low-sensitivity region.

5. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 1, characterized in that, The acquisition of the battery aging characteristic prediction sequence, and the identification of the necessary and preferred execution time windows for capacity recovery operations based on the aging characteristic prediction sequence, include: A battery aging characteristic prediction sequence is obtained, including capacity decay rate, internal resistance growth rate and polarization voltage change rate. The aging characteristic prediction sequence is generated by using a time series prediction model based on historical monitoring data of the battery management system. Based on the aging characteristic prediction sequence, the time points when the capacity decay rate exceeds the decay threshold and the time points when the internal resistance growth rate exceeds the internal resistance threshold are analyzed to identify the necessary time window for capacity recovery operation. The decay threshold ranges from 15% to 25% of the battery capacity decay ratio relative to the initial capacity, and the internal resistance growth rate threshold ranges from 30% to 50% of the battery internal resistance growth ratio relative to the initial internal resistance. Simultaneously, analyze the time points when the capacity decay rate approaches the decay threshold to identify the optimal execution time window for capacity recovery; A first flexibility weight of one is assigned to the mandatory execution time window, and a second flexibility weight of less than one is assigned to the preferred execution time window. The value of the second flexibility weight ranges from 0.5 to 0.

9. The smaller the difference between the capacity decay rate and the decay threshold, the larger the value of the second flexibility weight is, and a recovery time flexibility label is generated.

6. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 1, characterized in that, The process of generating a scene set using the Monte Carlo sampling method and then filtering the scene set using a scene reduction algorithm to obtain a representative scene set includes: Before sampling, the ambient temperature and power grid price were normalized to mean based on the range, so that parameters with different dimensions and numerical ranges were uniformly mapped to the same numerical range. Random sampling is performed based on the probability distribution of ambient temperature and power grid price after the uncertainty parameters are normalized. Each sampling generates a scenario, and each scenario corresponds to a specific combination of uncertain parameters, including the ambient temperature sequence and power grid price sequence under that scenario. A scene reduction algorithm is used to filter the scene set. The similarity between scenes is calculated in the normalized parameter space. Representative scenes with large probability weights and significant differences are retained. Similar scenes are merged and their probability weights are accumulated. The similarity between the scenes is calculated using Euclidean distance. The similarity between two scenes is obtained by taking the square root of the sum of the squares of the normalized temperature difference and the squares of the normalized price difference at each time period. When the similarity is less than the similarity threshold, the scenes are considered similar. The value of the similarity threshold ranges from 5% to 15% of the diagonal length of the normalized parameter space. After the screening is completed, the parameters of the representative scenarios are denormalized to restore the original temperature and price values.

7. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 1, characterized in that, The construction of the joint decision objective function includes: The first sub-objective is to maximize the revenue from grid ancillary services. The total service cost obtained in response to grid dispatch within the future time window is calculated by summing the products of the dispatch decision variables for each time period, the provided power, and the grid price. The second sub-objective is to maximize the value of the battery throughout its entire life cycle. The battery health value is obtained by calculating the product of the capacity decay reduction caused by the capacity recovery operation and the battery replacement cost. The calculation method is the sum of the decision variables for performing capacity recovery in each time period and the product of the capacity decay reduction and the battery replacement cost in that time period. The two sub-targets are normalized by using a mean normalization method based on the range to map them to the same numerical range. The two normalized sub-objectives are weighted and fused together. The joint decision objective function is the product of the first weight coefficient and the normalized grid ancillary service revenue plus the product of the second weight coefficient and the normalized battery life cycle value. The sum of the first weight coefficient and the second weight coefficient is one. The first weighting coefficient ranges from 0.3 to 0.8, and the corresponding second weighting coefficient ranges from 0.2 to 0.

7. When focusing on economic benefits, the first weighting coefficient is greater than 0.6, and when focusing on battery health, the second weighting coefficient is greater than 0.

6.

8. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 7, characterized in that, The method for calculating the capacity attenuation is as follows: Based on the aging feature prediction sequence, a battery capacity prediction trajectory is generated when capacity recovery is not performed. Generate a predicted battery capacity trajectory after capacity recovery is performed during the time period; The amount of capacity decay reduced by the capacity recovery operation is the capacity difference between the two trajectories at the end of the evaluation time window; The method of using dynamic programming to solve the bi-objective optimization problem, which involves working backward from the end of the time window to find the optimal decision for each stage, includes: The future time window is divided into multiple decision-making stages, and the optimal decision for each stage is solved in reverse from the last stage. In each scenario, starting from the end of the time window, the objective function value of various decision choices is calculated under the premise of satisfying the recovery hard constraint. The decision choices include responding to grid dispatch or performing capacity recovery. If capacity recovery is performed, the temperature parameter is selected from the robust temperature parameter candidate set. The value function records the optimal cumulative return from the time period to the end of the current state. The state includes battery capacity and whether the recovery hard constraint is met. The recursion of the value function is that the optimal value of the current time period is equal to the maximum value of the sum of the immediate returns under each decision action and the value function of the next time period. When the decision action is to respond to grid dispatch, the immediate benefit is the product of the first weighting coefficient and the dispatch decision variables, the power supplied, and the grid price; when the decision action is to perform capacity restoration, the immediate benefit is the product of the second weighting coefficient and the capacity restoration decision variables, the capacity degradation, and the battery replacement cost. By working backwards to the current moment, we obtain the complete optimal decision sequence for this scenario.

9. The capacity recovery method based on the entire life cycle of a lithium-ion battery according to claim 1, characterized in that, The extraction of the execution instructions for the current time period from the robust optimal decision scheme includes: Extract the execution instructions for the current time period from the robust optimal decision scheme, and convert the decision variables into executable control instructions. For time period scheduling decision variables, when the decision variable takes the value of one, it is converted into an instruction to perform capacity restoration operation, and when the decision variable takes the value of zero, it is converted into an instruction to respond to grid scheduling. For temperature parameter decision variables, the temperature values ​​in the robust temperature parameter candidate set corresponding to the decision variables are converted into the target temperature setpoint of the thermal management system. The execution instructions include the scheduling decision for the current time period and the corresponding control parameters. If the execution capacity is restored, the control parameters include robust temperature control parameters and thermal management control margin. The thermal management control margin is determined based on the uncertainty range of the ambient temperature and the power capacity of the battery thermal management system. It is calculated as the smaller value of the uncertainty range of the ambient temperature and the power capacity of the battery thermal management system divided by the product of the battery specific heat capacity and the battery mass.

10. A capacity recovery system based on the entire life cycle of a lithium-ion battery, used to perform the method according to any one of claims 1-9, characterized in that, include: The uncertainty parameter set generation module is used to obtain power grid ancillary service demand forecast data, ambient temperature forecast data and corresponding uncertainty range within a future time window and generate an uncertainty parameter set. The robust temperature parameter generation module is used to acquire capacity recovery efficiency data of the battery at multiple temperature points and generate a continuous function curve of capacity recovery efficiency versus temperature. It performs sensitivity analysis on the function curve to identify low-sensitivity regions and evaluates and generates a robust temperature parameter candidate set through robustness evaluation indicators. The recovery time flexibility analysis module is used to obtain the battery aging characteristic prediction sequence and identify the necessary execution time window and the preferred execution time window for capacity recovery operation, and to assign flexibility weights to different time windows to generate recovery time flexibility labels. The scene tree generation module is used to generate a scene set based on an uncertainty parameter set using the Monte Carlo sampling method, to filter a representative scene set using a scene reduction algorithm, to assign probability weights to each representative scene and to generate a scene tree structure. The dual-objective optimization problem construction module is used to construct a joint decision objective function that includes maximizing the revenue of grid ancillary services and maximizing the value of the battery throughout its entire life cycle. It weights and fuses the two sub-objectives and sets constraints to generate an economic-health dual-objective optimization problem. The conditionally optimal decision-making module is used to solve the bi-objective optimization problem in each branch of the scene tree structure using dynamic programming algorithm and generate a set of conditionally optimal decision schemes for each scene. The robust optimization decision selection module is used to calculate the worst-case objective value of each scheme in the set of optimal decision schemes under all scenarios and select the scheme with the largest worst-case objective value as the robust optimal decision scheme. The execution instruction generation module is used to extract the execution instructions for the current period from the robust optimal decision scheme and output them to the energy storage system controller and the battery thermal management system. The rolling optimization trigger module is used to monitor the actual values ​​of grid demand, battery status and ambient temperature in real time. When the deviation between the actual value and the predicted value exceeds the deviation threshold, the probability weight of each scenario in the scenario tree structure is updated and rolling optimization is triggered.

Images