A charging and discharging strategy adaptive optimization method and system for an energy storage cabinet

Abstract

The application discloses a kind of energy storage cabinet-oriented charge-discharge strategy self-adaptive optimization method and system, it is related to battery charge-discharge control technical field, including acquisition load data, environmental data and battery operation data, form multiple-source operation data sequence according to control period;Based on multiple-source operation data sequence, load prediction model is established, and load prediction sequence is generated;Battery operation data is estimated, and state of charge and online available capacity are obtained, and are fused to form fusion health state with long-term available capacity;Estimate battery internal temperature in combination with battery operation data, environmental data and fusion health state, form comprehensive risk index;Fusion health state and battery internal temperature are mapped to basic charge-discharge capacity, and charge-discharge power and state of charge range are tightened according to comprehensive risk index, and form dynamic safe operation boundary;Under the constraint of dynamic safe operation boundary, execute rolling optimization, output first charge-discharge power instruction, and execute cyclically with data update.

Description

Technical Field

[0001] This invention relates to the field of battery charging and discharging control technology, specifically to an adaptive optimization method and system for charging and discharging strategies of energy storage cabinets. Background Technology

[0002] With the expansion of new energy grid connection and the gradual application of peak-valley electricity pricing mechanisms, energy storage cabinets have been widely used in industrial and commercial peak shaving and valley filling, demand control, backup power supply, and new energy consumption. Existing energy storage management technologies typically collect user load, battery state of charge, electricity price, and environmental parameters, and combine them with load forecasting, time-of-use pricing, and model predictive control to generate charge and discharge plans. At the same time, battery equivalent circuits, health state estimation, and thermal state analysis are also gradually being introduced into energy storage control to reduce the impact of deep charge and discharge, high-temperature operation, and frequent power changes on battery life.

[0003] Current energy storage cabinet charging and discharging control still largely relies on fixed power thresholds, fixed state of charge (SOC) ranges, or preset time periods, making it difficult to simultaneously adapt to user load fluctuations, battery capacity degradation, and changes in environmental conditions. While some methods incorporate load forecasting, the forecasting models remain unchanged long-term after offline training, and prediction biases tend to accumulate when production plans, seasonal temperatures, or electricity usage habits change. Regarding battery state assessment, online voltage and current estimation results are easily affected by sensor noise and operating condition disturbances, and long-term aging models may exhibit parameter drift. Relying on only one method makes it difficult to balance short-term responsiveness and long-term stability. Furthermore, existing methods typically limit charging and discharging power based on measured temperature or fixed temperature thresholds, failing to adequately consider the combined effects of cabinet heat dissipation, ambient temperature, wind speed, solar radiation, and battery health. Power and SOC boundaries cannot be continuously adjusted according to risk conditions. Some economic dispatch methods primarily optimize electricity purchase and sale costs, lacking a unified correlation between battery aging costs, operational risks, and safety boundaries. Therefore, it is difficult to continuously generate charging and discharging commands that match the current state under conditions of changing load forecasts, battery performance degradation, and environmental risks. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing energy storage cabinet charging and discharging strategy optimization methods have the following problems: load prediction models are difficult to adapt to changes in user electricity consumption patterns; online battery capacity estimation and long-term capacity estimation are difficult to balance real-time performance and stability; fixed charging and discharging power boundaries and state of charge boundaries are difficult to reflect changes in battery temperature, environmental conditions and health status; and how to comprehensively consider grid energy consumption costs, battery aging costs and operational risks under dynamic safe operation boundary constraints, and generate charging and discharging power commands that match the current operating state of the energy storage cabinet.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: an adaptive optimization method for charging and discharging strategies of energy storage cabinets, comprising: collecting load data, environmental data, and battery operation data; forming a multi-source operation data sequence according to a control cycle; establishing a load prediction model based on the multi-source operation data sequence to generate a load prediction sequence; performing state estimation on the battery operation data to obtain the state of charge and online available capacity, and merging it with the long-term available capacity to form a fused health state; estimating the battery internal temperature by combining battery operation data, environmental data, and the fused health state to form a comprehensive risk index; mapping the fused health state and battery internal temperature to the basic charging and discharging capacity, and tightening the range of charging and discharging power and state of charge according to the comprehensive risk index to form a dynamic safe operation boundary; performing rolling optimization under the constraints of the dynamic safe operation boundary, outputting the first charging and discharging power command, and cyclically executing it as the data is updated.

[0007] As a preferred embodiment of the adaptive optimization method for charging and discharging strategies of energy storage cabinets described in this invention, the process of forming a multi-source operation data sequence includes: collecting active power load on the user side as load data; collecting ambient temperature, ambient humidity, light intensity, and wind speed as environmental data; collecting battery voltage, battery current, and battery temperature as battery operation data; matching the load data, environmental data, and battery operation data to corresponding time intervals according to the control cycle; aggregating the sampled data within the same time interval; comparing the aggregated sampled data with the corresponding physical range and range of change, and discarding and compensating the sampled data that exceeds the corresponding range; and combining the aggregated, discarded, and compensated sampled data in sequence according to the control cycle to form a multi-source operation data sequence.

[0008] As a preferred embodiment of the adaptive optimization method for charging and discharging strategies of energy storage cabinets described in this invention, the load prediction model includes a prediction input layer, a time-series feature extraction layer, and a prediction output layer connected in sequence; continuous load data, corresponding time-period features, and ambient temperature are extracted from multi-source operating data sequences as training inputs, and the actual load of continuous control periods after the training input is used as training labels to construct a training sample set; the prediction input layer performs time-series combination of the training inputs; the time-series feature extraction layer extracts the load change relationship between adjacent control periods; the prediction output layer maps the extraction results to the load prediction values ​​of continuous control periods; the load prediction values ​​and training labels are compared, and the parameters of the time-series feature extraction layer and the prediction output layer are adjusted according to the prediction deviation until the prediction deviation meets the convergence condition; within the current control cycle, the latest continuous load data, corresponding time-period features, and ambient temperature are input into the trained load prediction model to generate an optimized load prediction sequence in the time domain; when the prediction deviation within the continuous control cycle meets the update condition, the newly added actual load and corresponding training input constitute the update sample to update the parameters of the load prediction model.

[0009] As a preferred embodiment of the adaptive optimization method for charging and discharging strategies for energy storage cabinets described in this invention, the process of obtaining the state of charge (SOC) and online available capacity includes: accumulating the product of battery current and sampling time to update the SOC, and correcting the SOC according to the deviation between the measured and predicted battery terminal voltage values; combining the correspondence between battery voltage, battery current, and open-circuit voltage with the SOC to recursively deduce the battery internal resistance and online available capacity, and characterizing the reliability of the online capacity estimation according to the fluctuation of the voltage prediction deviation; accumulating the battery charge and discharge, forming a long-term capacity decay amount according to the capacity decay relationship between the accumulated results and battery temperature, and subtracting the long-term capacity decay amount from the battery nominal capacity to obtain the long-term available capacity; adjusting the fusion ratio of online available capacity and long-term available capacity according to the relative magnitude of the online capacity estimation reliability and the long-term capacity estimation reliability to obtain the fused available capacity; and characterizing the fused health status by the ratio of the fused available capacity to the battery nominal capacity.

[0010] As a preferred embodiment of the adaptive optimization method for charging and discharging strategies for energy storage cabinets described in this invention, the process of forming a comprehensive risk index includes: generating battery heat according to the heat conversion relationship between battery current and battery internal resistance; generating heat dissipation by combining the temperature difference between battery temperature and ambient temperature and wind speed, and generating radiant heat according to light intensity; recursively estimating battery temperature based on the heat balance relationship between battery heat generation, heat dissipation, and radiant heat, and then correcting it according to the deviation between the estimated temperature and the collected battery temperature to obtain the battery internal temperature; generating a temperature risk value according to the degree to which the battery internal temperature deviates from the safe temperature range, generating a humidity risk value according to the degree to which the ambient humidity deviates from the safe humidity range, and generating a health risk value according to the degree to which the fused health state decreases relative to the health state benchmark; and normalizing and weighting the temperature risk value, humidity risk value, and health risk value to obtain a comprehensive risk index.

[0011] As a preferred embodiment of the adaptive optimization method for charging and discharging strategies for energy storage cabinets described in this invention, the process of forming a dynamic safe operating boundary includes: establishing a mapping relationship between the fused health state, the battery internal temperature, and the basic charging and discharging capacity; the basic charging and discharging capacity includes a basic charging power limit and a basic discharging power limit; substituting the current fused health state and the current battery internal temperature into the mapping relationship, and obtaining the basic charging power limit and the basic discharging power limit according to the charging and discharging capacity change relationship within the corresponding interval; synchronously reducing the basic charging power limit and the basic discharging power limit according to the increase in the comprehensive risk index to obtain the safe charging power limit and the safe discharging power limit; increasing the basic state of charge limit and decreasing the basic state of charge upper limit according to the increase in the comprehensive risk index to obtain the safe state of charge lower limit and the safe state of charge upper limit; and combining the safe charging power limit, the safe discharging power limit, the safe state of charge lower limit, and the safe state of charge upper limit to form a dynamic safe operating boundary.

[0012] As a preferred embodiment of the adaptive optimization method for charging and discharging strategies of energy storage cabinets described in this invention, the following steps are included: Performing rolling optimization under dynamic safe operation boundary constraints involves dividing the optimization time domain into continuous control periods and setting the charging and discharging power of each control period as the power to be optimized; forming the grid interaction power for each control period according to the power balance relationship between the load forecast sequence and the power to be optimized, and reading the corresponding electricity price data for each control period; combining the grid interaction power and electricity price data to form the grid energy consumption cost; combining the power to be optimized, the integrated health status, and the battery internal temperature to form the battery aging cost; and combining the power to be optimized and the comprehensive risk index to form the operation cost. The optimization target is formed by accumulating the grid energy consumption cost, battery aging cost, and operational risk cost for each control period; constraining the power to be optimized with safe charging power limits and safe discharging power limits, recursively deducing the state of charge (SOC) based on the battery energy change caused by the SOC, and limiting the SOC between the safe SOC lower limit and the safe SOC upper limit; obtaining the charging and discharging power sequence with the lowest optimization target under the condition that the grid interaction power meets the grid interaction capacity range; executing the first charging and discharging power command corresponding to the current control period in the charging and discharging power sequence, and performing optimization again based on the updated multi-source operation data sequence after the current control period ends.

[0013] As a preferred embodiment of the adaptive optimization system for charging and discharging strategies for energy storage cabinets described in this invention, the system includes: a multi-source data module, a load prediction module, a state fusion module, a risk assessment module, a safety boundary module, and a rolling optimization module. The multi-source data module collects load data, environmental data, and battery operation data, forming a multi-source operation data sequence according to a control cycle. The load prediction module establishes a load prediction model based on the multi-source operation data sequence and generates a load prediction sequence. The state fusion module estimates the state of charge and online available capacity from the battery operation data, and merges it with the long-term available capacity to form a fused health state. The risk assessment module estimates the battery's internal temperature by combining battery operation data, environmental data, and the fused health state, forming a comprehensive risk index. The safety boundary module maps the fused health state and battery internal temperature to the basic charging and discharging capacity, tightens the range of charging and discharging power and state of charge according to the comprehensive risk index, forming a dynamic safe operation boundary. The rolling optimization module performs rolling optimization under the constraints of the dynamic safe operation boundary, outputs the first charging and discharging power command, and executes it cyclically as data is updated.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement an adaptive optimization method for charging and discharging strategies for energy storage cabinets.

[0015] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of an adaptive optimization method for a charging and discharging strategy for an energy storage cabinet.

[0016] The beneficial effects of this invention are as follows: The adaptive optimization method for charging and discharging strategies of energy storage cabinets provided by this invention dynamically integrates online capacity estimation driven by voltage and current with long-term capacity estimation based on temperature and throughput according to their respective uncertainties. The integrated health status can simultaneously reflect recent changes in battery operation and long-term degradation trends, reducing estimation drift caused by fixed weights or single capacity models. By mapping the integrated health status and battery internal temperature to the basic charge-discharge rate, and then using a comprehensive risk index formed by temperature, humidity, and health risks to simultaneously tighten the power boundary and state of charge boundary, the risk quantification results are directly converted into executable constraints, avoiding alarms without changing the control range. Rolling optimization integrates grid energy consumption, battery aging, and operational risks within the dynamic boundary. Each cycle executes only the first instruction and recalculates, allowing the charging and discharging strategy to be continuously updated with load forecasts, thermal state, and capacity state changes, reducing overcharging and discharging and mismatch scheduling while maintaining safety constraints. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 The present invention provides an overall flowchart of an adaptive optimization method for charging and discharging strategies of energy storage cabinets.

[0019] Figure 2 A schematic diagram of a computer device provided by the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0021] Reference Figure 1 As an embodiment of the present invention, an adaptive optimization method for charging and discharging strategies of energy storage cabinets is provided, comprising: S1: Collect load data, environmental data, and battery operation data, and form a multi-source operation data sequence according to the control cycle.

[0022] Furthermore, the process of forming a multi-source operational data sequence includes collecting active load data from the user side as load data; collecting ambient temperature, humidity, light intensity, and wind speed data as environmental data; collecting battery voltage, current, and temperature data as battery operational data; matching the load data, environmental data, and battery operational data to corresponding time intervals according to the control cycle; aggregating the sampled data within the same time interval; comparing the aggregated sampled data with the corresponding physical range and range of change, and eliminating and compensating for sampled data that exceeds the corresponding range; and combining the aggregated, eliminated, and compensated sampled data in sequence according to the control cycle to form a multi-source operational data sequence.

[0023] It should be noted that one specific scheme for forming a multi-source operational data sequence includes, in this embodiment, using industrial and commercial energy storage cabinets as the implementation object, setting the control cycle of the charging and discharging strategy to 15 minutes. Load forecasting, electricity price reading, and rolling optimization all use the same control cycle. Indicates the control cycle index, number 1 Each control cycle starts from time 1. Begin, lasting 15 minutes. Indicates the first The start time of each control cycle.

[0024] The load data represents the active load on the user side, collected through user-side energy metering devices, with a preferred sampling interval of 1 minute. Based on the sampling timestamp, load sample values ​​within the same control cycle are matched to corresponding time intervals, and the arithmetic mean of the load sample values ​​that pass validity verification is taken to obtain the [number of values]. Load data for each control cycle The unit is kilowatt.

[0025] Environmental data includes ambient temperature Ambient humidity Light intensity and wind speed The data are collected by temperature sensors, humidity sensors, light intensity sensors, and wind speed sensors located around the energy storage cabinet, with a preferred sampling interval of 10 seconds. Within the same control cycle, the median of the effective sampled values ​​for each type of environmental data is taken as the environmental data for the corresponding control cycle, in order to reduce the impact of gusts, short-term obstruction, and isolated sensor pulses on the representative values ​​of the environmental data.

[0026] Battery operating data includes battery terminal voltage Battery current and battery temperature , This represents the sampling index of battery operating data, with the sampling interval denoted as . In this embodiment, 1 second is preferred. The discharge current is uniformly defined as positive and the charging current as negative.

[0027] When multiple battery temperature measuring points are set in the energy storage cabinet, the first Battery temperature at each sampling time Take the arithmetic mean of all valid temperature measurement points. Average measured temperature over one control cycle Take all valid values ​​within the corresponding control period. The arithmetic mean.

[0028] The second-level battery terminal voltage, battery current, and battery temperature sequences are retained within the corresponding control cycle, without compressing all battery operating data into a 15-minute average. The aggregated data from the control cycle is used for load forecasting and rolling optimization, while the second-level battery operating data is used for state-of-charge estimation, online available capacity estimation, and battery internal temperature extrapolation.

[0029] Battery nominal capacity The relationship between open-circuit voltage and state of charge, the initial value of ohmic internal resistance, and polarization resistance. and polarization capacitors These battery characteristic parameters are pre-stored. They are derived from battery factory test data, energy storage cabinet commissioning data, and parameter tables stored in the battery management device.

[0030] The sampled data underwent physical range verification and variation range verification sequentially. The physical range included ambient temperature -40℃~85℃, humidity 0%RH~100%RH, light intensity 0W / m²~2000W / m², wind speed 0m / s~60m / s, and battery temperature -40℃~125℃; the battery terminal voltage was 0.9. ~1.1 The battery current is -1.1. ~1.1 The load range is determined by the meter's range. Values ​​outside the range are marked as invalid. and These represent the maximum and minimum permissible voltages, respectively. and These represent the maximum allowable charging current and the maximum allowable discharging current, respectively.

[0031] Variation range verification is performed on sampled values ​​that have passed physical range verification. The five most recent valid sampled values ​​before the current sampled value are selected, and the mean and standard deviation of the valid sampled values ​​are calculated. If the deviation between the current sampled value and the mean exceeds three times the standard deviation, the current sampled value is marked as a sudden change sampled value to be verified.

[0032] Two consecutive samples following the abrupt change in the sampled value to be verified are subjected to continuity checks. When the latter two samples return to the range of variation of the valid samples before the abrupt change, the lower limit of the range is obtained by subtracting three times the effective standard deviation from the mean of the valid samples, and the upper limit of the range is obtained by adding three times the effective standard deviation to the mean. The effective standard deviation is the larger of the actual standard deviation and the sensor resolution. If two consecutive samples following the abrupt change in the sampled value to be verified both fall within the range of variation, it is determined to be a pulse anomaly; if two consecutive samples are located on the same side of the range of variation, it is determined to be a continuous operating condition change, and no anomaly is rejected.

[0033] If there are fewer than two valid historical sample values ​​that can be used to calculate the standard deviation, the range of variation verification is not performed, and only the physical range verification result is retained; if the calculated standard deviation is lower than the corresponding sensor resolution, the sensor resolution is used as the minimum judgment criterion for range of variation verification.

[0034] When a single sampling point is missing, linear interpolation is performed using the two valid sampling values ​​before and after the missing location. When the continuous missing time does not exceed one control cycle, the most recent valid sampling value before the missing location is used for compensation. Abnormal data is processed in the order of first marking it as invalid, then removing it, and then compensating for it.

[0035] When all environmental data of a certain type are invalid within a certain control cycle, compensation is performed using the corresponding environmental data from the previous control cycle. When there are no valid sampled values ​​for the user-side active load, battery terminal voltage, or battery current throughout the entire control cycle, the current control cycle is not added to the training samples of the load prediction model, no new charging / discharging power command is generated, and the current charging / discharging power command is set to zero until the corresponding data is recovered.

[0036] After completing time matching, aggregation, anomaly removal, and missing data compensation, the first... Multi-source operation data records for each control cycle include load data. Ambient temperature Ambient humidity Light intensity Wind speed Average measured temperature This includes the battery terminal voltage, battery current, and battery temperature sequences within the corresponding control cycle. The multi-source operating data records are arranged in chronological order according to the control cycle to form a multi-source operating data sequence.

[0037] S2: Establish a load forecasting model based on multi-source operational data sequences and generate load forecasting sequences.

[0038] Furthermore, the load forecasting model comprises a forecasting input layer, a time-series feature extraction layer, and a forecasting output layer connected in sequence. Continuous load data, corresponding time-period features, and ambient temperature are extracted from multi-source operational data sequences as training inputs. The actual load during consecutive control periods following the training input is used as the training label to construct a training sample set. The forecasting input layer performs time-series combination of the training inputs. The time-series feature extraction layer extracts the load change relationship between adjacent control periods. The forecasting output layer maps the extracted results to the load forecast values ​​for consecutive control periods. The load forecast values ​​and training labels are compared, and the parameters of the time-series feature extraction layer and the forecasting output layer are adjusted according to the forecast deviation until the forecast deviation meets the convergence condition. Within the current control cycle, the latest continuous load data, corresponding time-period features, and ambient temperature are input into the trained load forecasting model to generate an optimized load forecasting sequence in the time domain. When the forecast deviation within the consecutive control cycle meets the update condition, the newly added actual load and corresponding training input constitute the update sample to update the parameters of the load forecasting model.

[0039] It should be noted that one specific scheme for establishing a load forecasting model includes a forecasting input layer, a time-series feature extraction layer, and a forecasting output layer connected sequentially. In this embodiment, a two-layer long short-term memory network is used to construct the time-series feature extraction layer, with 64 hidden units in each layer, and a fully connected structure is used to construct the forecasting output layer.

[0040] Ideally, at least 90 days of historical load data, corresponding time-period characteristics, and ambient temperature should be extracted from multi-source operational data sequences to construct training samples. With a control period of 15 minutes, the past 24 hours contain 96 control periods; the input window length should be... Set as The prediction time domain is set to the next 24 hours, and the prediction length is... Set as .

[0041] Each training sample includes user-side active load for 96 consecutive completed control periods, the corresponding ambient temperature for each control period, and time period characteristics. Time period characteristics include hour, day of the week, and holiday indicators. Hours and days of the week are encoded using one-hot encoding, while holidays are represented by 0 and 1. The training labels are the actual load for the 96 consecutive control periods after the input window ends.

[0042] Load data are based on the load mean of the training sample set. and load standard deviation Standardization was performed, with the ambient temperature set according to the average temperature of the training sample set. and temperature standard deviation Standardize the input value. Standardization involves subtracting the mean from the input value and then dividing by the standard deviation.

[0043] When the load standard deviation is less than 0.1% of the rated power of the energy storage cabinet, use 0.1% of the rated power of the energy storage cabinet instead of the load standard deviation; when the ambient temperature standard deviation is less than the temperature sensor resolution, use the temperature sensor resolution instead of the ambient temperature standard deviation. After training is completed, , , as well as The parameters are saved together with the load forecasting model parameters, and the same standardized parameters are used in the forecasting phase and the parameter update phase.

[0044] The prediction input layer processes training samples according to the control time period sequence. Each input time period includes a standardized load value, a standardized ambient temperature value, a 24-dimensional hourly one-hot code, a 7-dimensional weekday one-hot code, and a 1-dimensional holiday marker, forming a total of 34-dimensional input feature vectors. The prediction input layer uses a 64x34 input weight matrix to perform a linear transformation on the 34-dimensional input feature vectors and superimposes it with a 64-dimensional input bias vector to obtain the 64-dimensional first-layer temporal input.

[0045] Both the first and second layers of the Long Short-Term Memory (LSTM) network have 64 hidden units. For any input time interval in the first layer, the current 64-dimensional temporal input is combined with the 64-dimensional hidden state of the previous input time interval to form 128-dimensional data. This 128-dimensional data is then processed using four sets of weight and bias parameters to obtain a 64-dimensional forget gate, a 64-dimensional input gate, a 64-dimensional output gate, and a 64-dimensional candidate state. The forget gate, input gate, and output gate are processed by the Sigmoid function, ensuring each element is between 0 and 1. The candidate state is processed by the hyperbolic tangent function, ensuring each element is between -1 and 1. The forget gate is element-wise multiplied by the cell state of the previous input time interval, and the input gate is element-wise multiplied by the candidate state. These two results are then added together to obtain the current cell state. The current cell state is then processed by the hyperbolic tangent function and element-wise multiplied by the output gate to obtain the current hidden state.

[0046] The first layer outputs a 64-dimensional hidden state at each input time interval. The second layer takes the 64-dimensional hidden state output by the first layer at the same time interval as its current input and forms the second layer's 64-dimensional cell state and 64-dimensional hidden state according to the same gating operation process. At the beginning of each training sample processing, the hidden states and cell states of both layers of the Long Short-Term Memory network are set to 64-dimensional zero vectors, and are recursively processed in the order of the 1st to the 96th input time interval.

[0047] The prediction output layer reads the 64-dimensional hidden state formed by the second layer in the 96th input time period, performs a fully connected transformation using a 96x64 output weight matrix, and superimposes a 96-dimensional output bias vector to obtain 96 standardized load forecast values. The first output value corresponds to the first control time period in the prediction time domain, and so on up to the 96th control time period. Each standardized load forecast value is multiplied by... Then add The load forecast values ​​in kilowatts are obtained and arranged in the order of the forecast periods to form a load forecast sequence. .

[0048] In the At the start of the first control cycle, the input to the load forecasting model is the first... The control cycle up to the first The data for the control cycle, the 0th forecast period in the load forecast sequence corresponds to the 1st control cycle. The first control cycle corresponds to the first forecast period. One control cycle. element Indicates the optimization of the first time domain Load forecast values ​​for each forecast period; This represents the control period index within the prediction time domain, with a value range of 0 to 95; Indicates the first In the load forecast sequence generated at the start of the control cycle, the first... Load forecast values ​​for each forecast period.

[0049] The training process uses mean squared error to evaluate the deviation between the predicted load value and the training label. Specifically, the square of the difference between the predicted load value and the actual load value is calculated for each of the 96 prediction periods, and then the average value is taken for all training samples and all prediction periods. The Adam optimization method is used to adjust the weights and biases in the prediction input layer, the two-layer long short-term memory network, and the prediction output layer according to the mean squared error.

[0050] The initial learning rate is preferably set to 0.001, the batch size to 128, and the maximum number of training epochs to 500. When the mean squared error of the validation samples decreases by less than [a certain value] for 10 consecutive epochs... When the prediction error meets the convergence condition, training is stopped, and the load prediction model parameters corresponding to the lowest verification error are saved.

[0051] At the end of each control cycle, the load value for the first predicted period generated at the beginning of the current control cycle is used. Actual load data in the same control cycle The comparison is performed. The single-step prediction error is taken as the absolute value of the difference between the two, and then divided by the larger of the actual load data and 1% of the rated power of the energy storage cabinet.

[0052] When the single-step prediction error of three consecutive control cycles exceeds 15%, the three control cycles are used as the end of the label window. 96 consecutive actual load values ​​are extracted from the historical multi-source operation data sequence as training labels. 96 consecutive load data, time period characteristics, and ambient temperature are then extracted as input to the load prediction model, thus forming three new sliding window training samples. If the continuous data required for the new samples is insufficient, this incremental training is not performed.

[0053] After incremental training is completed, the average prediction error before and after the update is compared using the same updated samples. If the average prediction error after the update decreases, the updated parameters are saved; if the average prediction error after the update does not decrease, the load prediction model parameters before the update are retained.

[0054] S3: Perform state estimation on battery operating data to obtain state of charge and online available capacity, and merge it with long-term available capacity to form a merged health state.

[0055] Furthermore, the process of obtaining the state of charge (SOC) and online available capacity includes: accumulating the product of battery current and sampling time to update the SOC; correcting the SOC based on the deviation between the measured and predicted battery terminal voltage values; recursively estimating the battery internal resistance and online available capacity by combining the correspondence between battery voltage, battery current, open-circuit voltage, and SOC; characterizing the reliability of the online capacity estimation based on the fluctuation of the voltage prediction deviation; accumulating the battery charge and discharge amounts, forming a long-term capacity decay amount based on the capacity decay relationship between the accumulated results and battery temperature, and subtracting the long-term capacity decay amount from the battery's nominal capacity to obtain the long-term available capacity; adjusting the fusion ratio of online available capacity and long-term available capacity based on the relative magnitude of the online capacity estimation reliability and the long-term capacity estimation reliability to obtain the fused available capacity; and characterizing the fused health status by the ratio of the fused available capacity to the battery's nominal capacity.

[0056] It should be noted that one specific approach to obtaining the state of charge (SOC) and online available capacity includes using load forecasting sequences to reflect future changes in electricity demand. However, whether the energy storage unit can perform charging and discharging according to the forecast results also depends on the battery's current SOC and actual available capacity. Considering that short-term voltage and current data are suitable for tracking capacity changes, while long-term throughput and temperature better characterize continuous aging trends, a joint estimation of battery state and parameters is further performed, and online capacity and long-term capacity are integrated to form a fused health state reflecting the current battery charging and discharging capabilities.

[0057] This embodiment uses a first-order Thevenin equivalent circuit model to describe the relationship between battery terminal voltage, state of charge, and polarization voltage, and employs a dual extended Kalman filter to estimate the state of charge, polarization voltage, ohmic internal resistance, and online available capacity.

[0058] The state variables of state filtering include the state of charge. and polarization voltage The parameters to be estimated in parameter filtering include ohmic internal resistance. and online available capacity .

[0059] The state of charge is recursively calculated using the product of the battery current and the sampling interval. Battery current. When the value is not less than zero, the decrease in state of charge is measured by multiplying the battery current and the sampling interval, dividing by 3600, and the discharge coulombic efficiency. and online available capacity The product of; when the battery current is less than zero, the charging coulombic efficiency will be... The battery current and sampling interval are multiplied together and then divided by 3600 and the online available capacity. Since the battery current is negative, the calculation result increases the state of charge.

[0060] Polarization state attenuation coefficient Pick The exponential function value. The polarization voltage at the next sampling moment is determined by the polarization voltage at the previous sampling moment and... The product of, and the current battery current, polarization resistance and The product is obtained by adding the products.

[0061] Predicted battery terminal voltage Take the open-circuit voltage corresponding to the current state of charge, and subtract the polarization voltage and the voltage drop across the ohmic internal resistance. The relationship between open-circuit voltage and state of charge was established using static test data for the corresponding battery model. In this embodiment, a fifth-order polynomial fitting was used.

[0062] When the energy storage cabinet is started for the first time, the initial state of charge (SOC) is taken from the value reported by the battery management device. When the battery meets the static calibration conditions, the initial SOC is obtained by looking up a table based on the current open-circuit voltage. The initial polarization voltage is taken as zero, and the online available capacity is taken as the battery's nominal capacity. The ohmic internal resistance is taken as the initial value in the parameter table, and the initial error covariance of the state filter and the parameter filter is set as a positive definite diagonal matrix.

[0063] At each sampling time, the state filter calculates the prior values ​​of the state of charge and polarization voltage based on the previous time's state of charge, polarization voltage, ohmic internal resistance, online available capacity, and current current, and forms the predicted value of the battery terminal voltage. Measured terminal voltage Compared with the predicted value The difference is used as the voltage prediction bias to correct the state of charge and polarization voltage. Parameter filtering uses the ohmic internal resistance and online available capacity as random walk parameters. The sensitivity of the terminal voltage to the ohmic internal resistance is obtained from the current current, and the sensitivity of the terminal voltage to the online available capacity is obtained by multiplying the sensitivity of the state of charge to the capacity by the slope of the open-circuit voltage curve. The voltage prediction bias is then used to correct the ohmic internal resistance and online available capacity.

[0064] Since the available online capacity is not directly reflected in the instantaneous algebraic relationship of the battery terminal voltage, parametric filtering first calculates the sensitivity of the predicted state of charge (SOC) to the change in available online capacity at the next sampling moment based on the recursive relationship of SOC. Then, it multiplies this sensitivity by the slope of the open-circuit voltage curve at the current SOC to form the parametric sensitivity of the predicted battery terminal voltage to the available online capacity. The parametric filtering then recursively corrects the ohmic resistance and available online capacity based on the ohmic resistance parameter sensitivity, the available online capacity parameter sensitivity, the parameter estimation covariance, and the voltage prediction bias.

[0065] To establish a numerical correlation between the fluctuation of voltage prediction deviation and the reliability of online capacity estimation, the variance of all effective voltage prediction deviations within the current control cycle is calculated at the end of each control cycle. When the number of effective voltage prediction deviations is less than two, the measurement noise variance obtained from the static calibration of the voltage sensor is used. The larger of the voltage prediction deviation variance and the sensor measurement noise variance is used as the measurement noise variance for parameter filtering.

[0066] The diagonal elements of the parameter estimation covariance matrix corresponding to the online available capacity are used as the online capacity estimation variance. When voltage prediction deviation fluctuations increase, the variance of online capacity estimation increases, and the reliability of online capacity estimation decreases; when voltage prediction deviation fluctuations decrease, the variance of online capacity estimation decreases, and the reliability of online capacity estimation increases.

[0067] set up For the first The last valid battery sampling time within each control cycle is taken. As the first Online available capacity per control cycle And take the capacity variance in the corresponding parameter estimated covariance matrix as .

[0068] For the The product of the absolute value of the battery current and the sampling interval within each control cycle is accumulated, and then divided by 3600 to obtain the newly added throughput capacity. Add the newly added throughput to the cumulative throughput of the previous control cycle. Get the current cumulative throughput power .

[0069] For newly commissioned batteries, the initial cumulative throughput capacity and initial cumulative capacity decay ratio All values ​​are set to zero. For batteries already in operation, the initial cumulative throughput is read from the historical throughput records stored in the battery management device, and the initial cumulative capacity decay ratio is determined based on the capacity calibration results. If historical throughput records are missing, the initial capacity calibration results are used as the initial value of the long-term usable capacity, and the initial cumulative capacity decay ratio is determined based on the ratio between the long-term usable capacity and the battery's nominal capacity.

[0070] To prevent the temperature of the current control cycle from affecting all historical throughput and to maintain the irreversibility of battery capacity aging, the additional capacity degradation rate in the current control cycle is calculated using the following formula: in, Indicates the first The percentage of capacity decay added per control cycle; Indicates the factor prior to the exponent; Indicates activation energy; Represents the ideal gas constant; Indicates the first Average measured temperature over one control cycle; The conversion constant for converting Celsius temperature to Kelvin absolute temperature; This indicates the reference throughput power used for fitting the aging data; This represents the power-law factor of throughput. The throughput is normalized by referencing the throughput, therefore both the pre-exponential factor and the rate of capacity decay are dimensionless. , , as well as The data was obtained by fitting the cyclic aging data of the corresponding battery model under different temperature conditions.

[0071] Current cumulative capacity degradation ratio Take the cumulative capacity decay ratio of the previous control cycle The sum of the non-negative new capacity attenuation ratios; if the new capacity attenuation ratio is less than zero, it is treated as zero; if the cumulative result is greater than 1, it is treated as 1. Long-term available capacity. Take the battery's nominal capacity and The product of.

[0072] Using at least two sets of offline capacity calibration data, calculate the difference between the calculated long-term available capacity and the measured available capacity at the same calibration time, and calculate the sample variance of each difference relative to the mean of the differences, which is used as the long-term capacity estimation variance. When calibration data for the corresponding model is missing, the standard deviation of long-term capacity estimation is selected within the range of 1% to 5% of the nominal battery capacity, and in this embodiment, 2% is preferred; after obtaining the offline capacity calibration data for the corresponding model, the initial standard deviation is replaced with the actual residual statistical results.

[0073] Online available capacity and long-term available capacity are dynamically fused according to the two types of estimated variances, as follows: in, Indicates the available capacity for convergence; Indicates the available online capacity; Indicates long-term available capacity; and These represent the variance of online capacity estimation and the variance of long-term capacity estimation, respectively.

[0074] As the variance of online capacity estimation increases, the proportion of long-term available capacity in the fusion result increases; conversely, as the variance of long-term capacity estimation increases, the proportion of online available capacity in the fusion result also increases. The lower bounds for both online capacity estimation variance and long-term capacity estimation variance are set to... This avoids the situation where the variances of two estimates are both zero, resulting in a zero denominator in the fusion calculation.

[0075] Integrated health status Take the available capacity of the fusion With the battery's nominal capacity The ratio of online available capacity, long-term available capacity, and combined available capacity is limited to between 0 and the nominal battery capacity, while the combined health status is limited to between 0 and 1.

[0076] It should also be noted that a first-order Thevenin equivalent circuit and dual extended Kalman filters are used to estimate the state of charge, polarization voltage, ohmic internal resistance, and online available capacity, so that capacity estimation no longer relies solely on cumulative current or a single capacity test. Voltage prediction bias corrects for battery state on the one hand, and corrects for slowly varying parameters by adjusting the sensitivity of terminal voltage to ohmic internal resistance and online available capacity parameters on the other hand, with the capacity estimation covariance reflecting the uncertainty of online results. Long-term capacity is calculated based on the cumulative decay of new throughput and corresponding temperature in each control cycle, with the current temperature only affecting the current aging, avoiding capacity rebound caused by re-evaluating historical throughput at high temperatures under low-temperature conditions. Online capacity and long-term capacity are not used in a fixed ratio, but are dynamically fused based on their respective estimation variances: when short-term voltage data is stable, the influence of online results is increased; when online disturbances are large, the benchmark role of long-term capacity is enhanced. The resulting fused available capacity takes into account both rapid tracking and long-term trends, and is further converted into a fused healthy state, reducing capacity judgment bias caused by noise, changes in operating conditions, or model drift in a single estimation path.

[0077] S4: Combine battery operation data, environmental data, and integrated health status to estimate the battery's internal temperature and form a comprehensive risk index.

[0078] Furthermore, the process of forming the comprehensive risk index includes: generating battery heat based on the heat conversion relationship between battery current and battery internal resistance; generating heat dissipation by combining the temperature difference between battery temperature and ambient temperature and wind speed, and generating radiant heat based on light intensity; recursively estimating battery temperature based on the heat balance relationship between battery heat generation, heat dissipation, and radiant heat, and then correcting for the deviation between the estimated temperature and the collected battery temperature to obtain the battery internal temperature; generating a temperature risk value based on the degree to which the battery internal temperature deviates from the safe temperature range, generating a humidity risk value based on the degree to which the ambient humidity deviates from the safe humidity range, and generating a health risk value based on the degree to which the fused health status decreases relative to the health status benchmark; and then normalizing and weighting the temperature risk value, humidity risk value, and health risk value to obtain the comprehensive risk index.

[0079] It should be noted that one specific scheme for forming a comprehensive risk index involves treating the battery pack and supporting structure in the energy storage cabinet as a single heat capacity, using the average internal temperature of the battery as the temperature state quantity. The battery's heat generation power is calculated based on the battery current and ohmic internal resistance; the cabinet's heat dissipation power is calculated based on the battery's internal temperature, ambient temperature, wind speed, and the cabinet's heat dissipation area; and the solar radiation input heat power is calculated based on light intensity, effective radiation area, and solar radiation absorptivity. Then, by subtracting the heat dissipation power from the heat generation power and adding the solar radiation input heat power, and combining this with the sampling interval and equivalent heat capacity, the predicted value of the battery's internal temperature at the next sampling time is recursively obtained.

[0080] No. The battery heat generation power at each sampling time is taken as the square of the battery current and the ohmic internal resistance. The product of solar radiation input heat power and solar irradiance intensity. Effective radiation area of ​​the cabinet and solar radiation absorption rate The product of.

[0081] Solar radiation absorptivity The value is obtained from the surface material parameters of the cabinet; in this embodiment, it is preferably 0.6. The effective radiation area is obtained based on the projected area of ​​the top and main light-receiving sides of the cabinet; in this embodiment, it is preferably [value missing]. .

[0082] The natural convection heat transfer coefficient is based on the results of a cabinet zero-wind-speed heat dissipation test. In this embodiment, the preferred coefficient is... The forced convection heat transfer coefficient is based on... Calculate wind speed The calculated forced convection heat transfer coefficient is expressed in watts per square meter (Kelvin), measured in meters per second.

[0083] The total heat transfer coefficient is obtained by adding the natural convection heat transfer coefficient and the forced convection heat transfer coefficient. The reciprocal of the product of the total heat transfer coefficient and the effective heat dissipation area of ​​the cabinet is used to obtain the equivalent thermal resistance. The effective heat dissipation area of ​​the cabinet is determined based on the geometric dimensions of the outer surface of the cabinet; in this embodiment, a preferred value is [missing value]. .

[0084] Using the first-order forward Euler method to determine the heat dissipation equilibrium relationship, the prior predicted value of the battery's internal temperature is obtained, expressed as: in, This represents the prior prediction of the battery's internal temperature. This indicates the internal temperature of the battery after measurement correction at the previous sampling time; Indicates the battery operation data sampling interval; This represents the equivalent heat capacity of the battery pack and its supporting structure. This represents the equivalent thermal resistance.

[0085] The equivalent heat capacity is calibrated through a heating test of the energy storage cabinet, and the finite element simulation results are used as the calibration verification. In this embodiment, 5000 J / K is preferred.

[0086] The first-order Kalman filter corrects the battery internal temperature according to the following process: the temperature error variance at the previous sampling time is increased by the temperature recursion noise variance to obtain the current prior error variance; the prior error variance is divided by the sum of the prior error variance and the battery temperature measurement noise variance to obtain the temperature correction gain; the difference between the battery temperature measurement value and the temperature prior prediction value forms the temperature prediction bias; the temperature correction gain is multiplied by the temperature prediction bias, and then added to the temperature prior prediction value to obtain the corrected battery internal temperature; the difference between step 1 and the temperature correction gain is calculated, and the resulting difference is multiplied by the prior error variance to obtain the corrected temperature error variance.

[0087] The noise variance of the temperature recursion process is obtained by calculating the residual variance between the predicted temperature value obtained from the thermal balance relationship and the temperature test data. The noise variance of the battery temperature measurement is obtained by calculating the residual variance of the temperature sensor static test. In this embodiment, the standard deviation of the temperature recursion process noise is preferably set to 0.05℃, and the standard deviation of the battery temperature measurement noise is set to 0.3℃.

[0088] When the energy storage cabinet starts up, the arithmetic mean of all effective temperature measurement points is used as the initial internal temperature of the battery. The corrected temperature at the last valid sampling moment of each control cycle is recorded as the battery internal temperature. .

[0089] Temperature risk value The following rules apply to the calculation: When the internal temperature of the battery is below 0°C, the square of the internal temperature value divided by 5 is calculated, and the result is limited to no more than 1; when the internal temperature of the battery is between 0°C and 40°C, the temperature risk value is 0; when the internal temperature of the battery is above 40°C, the square of the difference between the internal temperature of the battery and 40°C divided by 15°C is calculated, and the result is limited to no more than 1.

[0090] Humidity risk value The following rules apply to the calculation: when the ambient humidity is not higher than 60%, the humidity risk value is 0; when the ambient humidity is higher than 60% but not higher than 95%, the square of the difference between the ambient humidity and 60% divided by 35 percentage points is calculated; when the ambient humidity is higher than 95%, the humidity risk value is 1.

[0091] The comprehensive risk index is formed by weighting temperature risk value, humidity risk value, and health risk value, and is expressed as follows: in, This represents the overall risk index; This means that the calculation result will be limited to between 0 and 1; , as well as These represent the weights of temperature risk, humidity risk, and health risk, respectively. All three weights are non-negative numbers between 0 and 1, and the sum of the three weights is 1. The weights are determined based on the analytic hierarchy process (AHP) and energy storage cabinet failure statistics; this embodiment preferably sets them as follows. , , .

[0092] S5: Based on the mapping of health status and battery internal temperature, the charging and discharging capabilities are integrated, and the charging and discharging power and state of charge range are tightened according to the comprehensive risk index to form a dynamic safe operating boundary.

[0093] Furthermore, the process of forming a dynamic safe operating boundary includes establishing a mapping relationship between the integrated health state, battery internal temperature, and basic charge / discharge capacity; the basic charge / discharge capacity includes basic charging power limits and basic discharging power limits; substituting the current integrated health state and current battery internal temperature into the mapping relationship, and obtaining the basic charging power limits and basic discharging power limits according to the charging / discharging capacity changes within the corresponding intervals; synchronously reducing the basic charging power limits and basic discharging power limits according to the increase in the comprehensive risk index to obtain the safe charging power limits and safe discharging power limits; increasing the basic state of charge lower limit and decreasing the basic state of charge upper limit according to the increase in the comprehensive risk index to obtain the safe state of charge lower limit and safe state of charge upper limit; and combining the safe charging power limits, safe discharging power limits, safe state of charge lower limit, and safe state of charge upper limit to form the dynamic safe operating boundary.

[0094] It should be noted that one specific scheme for forming a dynamic safe operating boundary includes the following: the battery's internal temperature and comprehensive risk index can characterize the current thermal state, environmental state, and degree of health degradation, but the risk assessment results themselves cannot be directly used as control commands for the energy storage converter. To convert the state assessment into executable operating limits, the basic charge and discharge capacity is further determined based on the integrated health state and battery internal temperature, and then the power range and state of charge range are tightened using the comprehensive risk index to form a dynamic safe operating boundary that changes with the battery state.

[0095] This embodiment establishes a two-dimensional mapping relationship between the integrated health state, battery internal temperature, and basic charge / discharge rate through charge / discharge rate tests on the corresponding battery model. During the test, the battery is adjusted to different integrated health state nodes and temperature nodes. After the corresponding temperature stabilizes, the charge rate or discharge rate is gradually increased, and the battery terminal voltage, battery current, and battery temperature are monitored. The maximum rate corresponding to the condition that does not exceed the voltage boundary, temperature boundary, and current boundary specified by the battery manufacturer is taken as the basic charge rate or basic discharge rate of the corresponding node.

[0096] The two-dimensional mapping table used in this embodiment is shown in Table 1. Vertical nodes in the table represent the fused health state, and horizontal nodes represent the battery's internal temperature. The fused health state is displayed as a percentage in the table and converted to a ratio of 0 to 1 when used in calculations. The first value in each table cell represents the base charge rate, and the second value represents the base discharge rate, both in C-rate. For example, when the fused health state is 100% and the battery's internal temperature is 25°C, both the base charge rate and the base discharge rate are 1.00C; when the fused health state is 80% and the battery's internal temperature is 0°C, the base charge rate is 0.28C and the base discharge rate is 0.80C.

[0097] Table 1 Basic Charge / Discharge Rate Mapping Table Table 1 presents a directly executable calibration result. Different battery models were re-obtained using the same rate testing method to acquire the corresponding node values. Table 1 reflects the decrease in the allowable charge / discharge rate of the battery when the fusion health status declines or the battery's internal temperature deviates from the suitable temperature range. When the current fusion health status and battery internal temperature coincide with a node in the mapping table, the base charge rate and base discharge rate in the corresponding cell are directly read; when located between adjacent nodes, bilinear interpolation is used.

[0098] The calculation process of bilinear interpolation is as follows: determine the health state nodes on both sides of the current fused health state. and And the temperature nodes on both sides of the current internal temperature of the battery. and Subtract the current fusion health status Divide by This yields the positional ratio of the current fused health state between two adjacent health state nodes. ; Subtract the current internal temperature of the battery Divide by This gives the ratio of the current internal temperature of the battery to the position between two adjacent temperature nodes. .

[0099] For both charging and discharging rates, a four-corner weighting is applied: the rate of nodes with lower health status and lower temperature is multiplied by [the factor]. The multiplier of nodes with lower health status and higher temperature is multiplied by The node with higher health status and lower temperature multiplier is multiplied by The node with higher health status and higher temperature is multiplied by The four results are added together to obtain the current base charge rate or base discharge rate.

[0100] When the fusion health status is higher than 1, it is treated as 1; when the fusion health status is between 0.60 and 1, it is retrieved and interpolated according to the mapping table. When the fusion health status is lower than 0.60, both the base charge rate and the base discharge rate are set to zero, and regular charging and discharging are stopped.

[0101] When the battery internal temperature is below -20℃ or above 60℃, the basic charge rate and basic discharge rate are both set to zero, and regular charging and discharging are not performed; when the battery internal temperature is between -20℃ and 60℃, the value is retrieved according to the mapping table or by bilinear interpolation.

[0102] The basic charging current limit is taken as the basic charging rate and the battery nominal capacity. The product of the two values, the basic discharge current limit is taken as the basic discharge rate and the battery nominal capacity. The product of.

[0103] No. The battery terminal voltage for each control cycle is taken from the last valid battery terminal voltage sample value within the corresponding control cycle. The basic charging power limit and the basic discharging power limit are respectively taken from the corresponding current limit and... The product of and divided by 1000 is converted to kilowatts.

[0104] The comprehensive risk index tightens both the charge / discharge power range and the state of charge range, as shown below: in, and These represent the basic charging power limit and the basic discharging power limit, respectively. and These represent the safe charging power limit and the safe discharging power limit, respectively. This represents the risk adjustment factor; and These represent the lower limit of the basic state of charge and the upper limit of the basic state of charge, respectively. This represents the maximum charge state boundary offset; and These represent the lower limit of the safe state of charge and the upper limit of the safe state of charge, respectively.

[0105] Risk Adjustment Coefficient The value of is between 0 and 1, and is determined through charging and discharging safety tests under high-risk conditions. In this embodiment, 0.8 is preferred, so that 20% of the basic charging and discharging power is retained when the comprehensive risk index is 1; when charging and discharging are required to be stopped under the highest risk conditions, the risk correction coefficient is set to 1.

[0106] The lower limit and upper limit of the basic state of charge (PSC) are determined based on the battery manufacturer's recommended operating range; in this embodiment, they are taken as 0.10 and 0.90, respectively. The maximum PSC boundary offset is determined through deep charge / discharge safety margin tests; in this embodiment, it is preferably taken as 0.08. The maximum PSC boundary offset is not less than zero and is less than half the difference between the upper and lower limits of the basic PSC, ensuring that the lower limit of the safe PSC is always lower than the upper limit of the safe PSC.

[0107] The safe charging power limit, the safe discharging power limit, the safe state of charge limit, and the safe state of charge upper limit together constitute the dynamic safe operating boundary.

[0108] It should also be noted that the base charge / discharge capacity is determined based on the position of the integrated health status and battery internal temperature in the two-dimensional rate mapping table, so that power limitation no longer relies solely on a fixed rated rate or a single temperature derating curve. The node values ​​in the mapping table are derived from rate tests for the corresponding battery models. When the health status declines, the allowed rate decreases synchronously with capacity decay; when the temperature deviates from the suitable range, the charging and discharging rates shrink according to the test results. When the current status coincides with a node, it is directly retrieved; when it is between four adjacent nodes, bilinear interpolation is used to reduce power boundary jumps caused by discrete rate switching. When the integrated health status is below the lower limit of the table or the temperature exceeds the calibrated range, the base rate is set to zero to avoid extrapolation outside the table in areas lacking test support. The comprehensive risk index further tightens the base charge / discharge power and simultaneously raises the lower limit of the state of charge and lowers the upper limit of the state of charge, thereby forming a dynamic safe operating boundary that simultaneously limits the instantaneous power and the usable capacity range, so that environmental and health risks not only participate in the evaluation but also directly change the executable charge / discharge range.

[0109] S6: Performs rolling optimization under dynamic safe operation boundary constraints, outputs the first charge / discharge power command, and executes it cyclically as data is updated.

[0110] Furthermore, under dynamic safe operation boundary constraints, rolling optimization includes dividing the optimization time domain into continuous control periods and setting the charging and discharging power of each control period as the power to be optimized; forming the grid interaction power for each control period according to the power balance relationship between the load forecast sequence and the power to be optimized, and reading the corresponding electricity price data for each control period; forming the grid energy consumption cost by combining the grid interaction power and electricity price data, forming the battery aging cost by combining the power to be optimized, the integrated health status, and the battery internal temperature, and forming the operation risk cost by combining the power to be optimized and the comprehensive risk index; accumulating the grid energy consumption cost, battery aging cost, and operation risk cost for each control period to form the optimization target; constraining the power to be optimized with safe charging power limits and safe discharging power limits, recursively deducing the state of charge according to the battery energy change caused by the power to be optimized, and limiting the state of charge to between the safe state of charge limit and the safe state of charge upper limit; obtaining the charging and discharging power sequence with the lowest optimization target under the condition that the grid interaction power meets the grid interaction capacity range; executing the first charging and discharging power command corresponding to the current control period in the charging and discharging power sequence, and performing optimization again based on the updated multi-source operation data sequence after the current control period ends.

[0111] It should be noted that one specific scheme for performing rolling optimization under dynamic safety operation boundary constraints includes the dynamic safety operation boundary defining the allowable charging and discharging power and state of charge range of the energy storage cabinet under the current state, but multiple feasible charging and discharging schemes still exist within the boundary. To meet safety constraints while also considering time-of-use pricing, battery aging, and operational risks, rolling optimization is further performed in the prediction time domain to obtain the charging and discharging power sequence, and only the first power command corresponding to the current control cycle is executed, ensuring that subsequent decisions are continuously updated with new operational data.

[0112] This embodiment employs model predictive control to perform rolling optimization, with one optimization calculation performed per control cycle. The optimization time domain is the next 24 hours, comprising 96 consecutive control periods, each with a length of 0.25 hours.

[0113] No. Within the first control cycle, the first The charge / discharge power to be optimized for each prediction period is denoted as . The discharge power is defined as positive and the charging power as negative. The charging and discharging power to be optimized represents the net output power on the AC side of the energy storage converter.

[0114] The purchase price and sales price of electricity are read from the pre-stored time-of-use electricity price table, the grid interaction capacity is read from the user's grid connection parameters, and the charging efficiency, discharging efficiency, battery purchase cost, battery nominal voltage, and full life-cycle energy parameters are read from the battery and energy storage converter specification data.

[0115] No. The grid interaction power for each forecast period is taken from the load forecast value. Subtract the charging and discharging power to be optimized A positive grid interaction power indicates that electricity is purchased from the grid, while a negative grid interaction power indicates that electricity is supplied to the grid.

[0116] When the grid interaction power is positive, the grid interaction power, the corresponding electricity purchase price, and the control period length are multiplied to form the grid energy consumption cost; when the grid interaction power is negative, the absolute value of the grid interaction power, the corresponding electricity sales price, and the control period length are multiplied to form the electricity sales revenue, which is deducted from the grid energy consumption cost.

[0117] Let the cumulative unidirectional discharge energy that the battery can release over its entire lifespan under reference temperature and reference rate conditions be denoted as . Total energy throughput for both charging and discharging The baseline aging cost per unit bidirectional throughput energy is calculated by dividing the battery purchase or replacement cost by [missing information]. The unit is yuan per kilowatt-hour.

[0118] The aging acceleration factor is calculated based on the combined health state and the battery's internal temperature. First, the Arrhenius temperature acceleration ratio of the current battery's internal absolute temperature relative to a 25°C reference temperature is calculated, specifically the activation energy. Divide by the ideal gas constant Multiply by the difference between the reciprocal of the reference absolute temperature and the reciprocal of the current absolute temperature, and then take the exponential function value; then calculate the ratio of the reference fusion health state of 0.95 to the current fusion health state, and take the health state acceleration index. The power is calculated as follows: if the current fusion health status is below 0.60, it is calculated as 0.60; the temperature acceleration ratio is multiplied by the health acceleration ratio and then compared with 1, and the larger value is taken as the aging acceleration factor.

[0119] Health Acceleration Index The aging acceleration factor is obtained by fitting cyclic aging data under different health states and is greater than zero; in this embodiment, it is preferably set to 3. At 25°C and with a fusion health state not lower than 0.95, the aging acceleration factor is set to 1; as the internal temperature of the battery increases or the fusion health state decreases, the aging acceleration factor increases. Within the same rolling optimization cycle, the aging acceleration factor is calculated based on the current fusion health state and the current internal temperature of the battery and remains unchanged throughout the optimization process.

[0120] The battery aging cost for each prediction period is calculated as the product of the baseline aging cost, the aging acceleration factor, the absolute value of the charge / discharge power to be optimized, and the length of the control period, with the result in yuan.

[0121] The operational risk cost for each forecast period is calculated as the product of the comprehensive risk index, the square of the power to be optimized for charging and discharging, the control period length, and the risk penalty weight. Risk penalty weight. Greater than zero, unit is yuan This allows the costs of operational risks, grid energy consumption, and battery aging to all be expressed in monetary terms. Risk penalty weights are determined through power regulation tests and economic benefit comparisons under different comprehensive risk index conditions.

[0122] The rolling optimization objective is expressed as: in, Indicates the first The optimization target value for each control cycle; , as well as They represent the first The costs of grid energy consumption, battery aging, and operational risks for each forecast period.

[0123] The power to be optimized for charging and discharging must not be lower than the negative value of the safe charging power limit, nor higher than the safe discharging power limit. The grid interaction power must not exceed the grid interaction capacity range specified by the user's grid connection parameters.

[0124] Current available energy is taken as the fusion available capacity Multiply the voltage by the nominal battery voltage and divide by 1000 to convert to kilowatt-hours. The first state of charge in the optimized time domain is the current state of charge.

[0125] When the power to be optimized is the discharge power, the product of the power to be optimized and the control period length is divided by the discharge efficiency and the currently available energy to obtain the decrease in state of charge (SOC), which is then deducted from the current SOC. When the power to be optimized is the charging power, the absolute value of the power to be optimized, the control period length, and the charging efficiency are multiplied, and then divided by the currently available energy to obtain the increase in SOC, which is then added to the current SOC. The SOC obtained recursively for each prediction period must not be lower than the lower limit of the safe SOC, nor higher than the upper limit of the safe SOC.

[0126] Rolling optimization with single net charge / discharge power As decision variables, charging and discharging commands for the same prediction period are not output separately. The absolute power value in battery aging cost and the segmented relationship of power purchase and sale in grid energy cost are equivalently processed using non-negative auxiliary variables commonly used in numerical optimization. The auxiliary variables only serve the solution process, and the final output is still the net charging and discharging power sequence.

[0127] When both the safe charging power limit and the safe discharging power limit are zero, the first charging / discharging power command of the current control cycle is directly set to zero, and non-zero power optimization is not performed. In other cases, a numerical optimization method using adaptive linear constraints, piecewise linear costs, and quadratic risk costs is used to solve the problem, resulting in a charging / discharging power sequence containing 96 charging / discharging power values.

[0128] Specifically, for each forecast period, non-negative charging power, non-negative discharging power, power purchase, power sale, and state of charge variables are set. Net charging and discharging power is calculated as discharging power minus charging power, and grid interaction power is calculated as power purchase minus power sale. The absolute value of power in the battery aging cost is converted into the sum of charging and discharging power, and the segmented cost of power purchase and sale is converted into power purchase cost minus power sale revenue. The state of charge is then recursively derived according to charging efficiency and discharging efficiency. The charging and discharging power boundaries, state of charge boundaries, grid interaction capacity, and power balance relationship all form linear constraints. The operational risk cost is retained as a quadratic term of the net charging and discharging power, thus forming a quadratic programming problem.

[0129] The quadratic programming problem is solved iteratively using the primal-dual interior-point method until the change in the objective function, the constraint residuals, and the dual residuals are all below a preset convergence threshold. In this embodiment, the convergence threshold is set to... The maximum number of iterations is set to 100. Since both charging and discharging power are factored into positive aging costs, and the purchase price of electricity is not lower than the sales price, simultaneous charging and discharging during the same prediction period will not reduce the optimization objective. After convergence, the discharging power is subtracted from the charging power to obtain a charging and discharging power sequence containing 96 net charging and discharging power values; if no feasible solution is obtained, the charging and discharging power command for the current control cycle is set to zero.

[0130] Only the first charge / discharge power value in the charge / discharge power sequence is used as the first charge / discharge power command for the current control period and sent to the energy storage converter. Before sending the command, the first charge / discharge power command is compared again with the safe charging power limit and the safe discharging power limit; if the value is negative and lower than the safe charging power limit, the first charge / discharge power command is limited to the negative of the safe charging power limit; if the value is higher than the safe discharging power limit, the first charge / discharge power command is limited to the safe discharging power limit.

[0131] After the current control period ends, the updated load data, environmental data, and battery operation data are collected to re-form a multi-source operation data sequence, re-generate a load prediction sequence, re-estimate the state of charge, online available capacity, and long-term available capacity, re-form a fused health state, re-estimate the battery internal temperature and comprehensive risk index, re-generate the dynamic safe operation boundary, and perform rolling optimization again, thereby forming a closed-loop charge and discharge control.

[0132] It should also be noted that rolling optimization is performed for the next 96 control periods within the dynamic safety operation boundary, incorporating grid energy consumption costs, battery aging costs, and operational risk costs into the same objective. This avoids economic dispatching relying solely on peak-valley electricity prices to allocate batteries while ignoring current health status and environmental risks. Load forecasts participate in the grid interaction power calculation for each period, integrating health status and the aging acceleration factor influenced by battery internal temperature. The comprehensive risk index is simultaneously incorporated into both risk costs and the safety boundary, ensuring that economic efficiency, lifespan consumption, and operational risks share common decision variables. Charging power, discharging power, power purchase, and power sale are transformed into a quadratic programming form through non-negative auxiliary variables, with power balance, grid capacity, state of charge, and dynamic boundaries forming directly solvable constraints. Only the first power command in the optimization sequence is executed each time. After the control cycle ends, data is re-collected and the solution is repeated, avoiding a one-time fixed execution of the 24-hour plan. This allows load deviations, temperature changes, and capacity estimation updates to be incorporated into new control decisions in the next control cycle.

[0133] One embodiment of the present invention provides an adaptive optimization system for charging and discharging strategies for energy storage cabinets, including a multi-source data module, a load forecasting module, a state fusion module, a risk assessment module, a safety boundary module, and a rolling optimization module.

[0134] The system comprises the following modules: a multi-source data module for collecting load data, environmental data, and battery operation data, forming a multi-source operation data sequence according to the control cycle; a load prediction module for establishing a load prediction model based on the multi-source operation data sequence and generating a load prediction sequence; a state fusion module for estimating the state of charge and online available capacity from the battery operation data, and fusing it with the long-term available capacity to form a fused health state; a risk assessment module for estimating the battery's internal temperature by combining battery operation data, environmental data, and the fused health state, forming a comprehensive risk index; a safety boundary module for mapping the fused health state and battery internal temperature to the basic charge / discharge capacity, tightening the range of charge / discharge power and state of charge according to the comprehensive risk index, forming a dynamic safety operation boundary; and a rolling optimization module for performing rolling optimization under the constraints of the dynamic safety operation boundary, outputting the first charge / discharge power command, and executing it cyclically as the data is updated.

[0135] Reference Figure 2This embodiment also provides a computer device applicable to the adaptive optimization method for charging and discharging strategies of energy storage cabinets, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the adaptive optimization method for charging and discharging strategies of energy storage cabinets as proposed in the above embodiment.

[0136] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0137] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the adaptive optimization method for charging and discharging strategies for energy storage cabinets as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

Claims

1. An adaptive optimization method for charging and discharging strategies of energy storage cabinets, characterized in that, include: Collect load data, environmental data, and battery operation data, and form a multi-source operation data sequence according to the control cycle; A load forecasting model is established based on multi-source operational data sequences to generate load forecasting sequences; State estimation is performed on battery operating data to obtain state of charge and online available capacity, which are then integrated with long-term available capacity to form a fused health state. By combining battery operation data, environmental data, and integrated health status, the internal temperature of the battery is estimated to form a comprehensive risk index. Based on the mapping of health status and battery internal temperature, the charging and discharging capabilities are integrated, and the range of charging and discharging power and state of charge is tightened according to the comprehensive risk index to form a dynamic safe operating boundary. Under dynamic safety operation boundary constraints, rolling optimization is performed, the first charge / discharge power command is output, and the command is executed cyclically as data is updated.

2. The adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in claim 1, characterized in that: The process of forming a multi-source operation data sequence includes collecting active load on the user side as load data; collecting ambient temperature, ambient humidity, light intensity, and wind speed as environmental data; and collecting battery voltage, battery current, and battery temperature as battery operation data. According to the control cycle, load data, environmental data, and battery operation data are matched to the corresponding time intervals, and the sampled data within the same time interval are aggregated. The aggregated sampled data is compared with the corresponding physical range and range of change, and sampled data that exceeds the corresponding range is removed and compensated. The sampled data, which are aggregated, eliminated, and compensated in sequence according to the control cycle, are combined to form a multi-source operational data sequence.

3. The adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in claim 2, characterized in that: The load prediction model consists of a prediction input layer, a time-series feature extraction layer, and a prediction output layer connected in sequence. Continuous load data, corresponding time period characteristics, and ambient temperature are extracted from multi-source operational data sequences as training inputs, and the actual load during continuous control periods after the training input is used as training labels to construct a training sample set. The prediction input layer performs temporal combination of the training input; The time-series feature extraction layer extracts the load change relationship between adjacent control periods; the prediction output layer maps the extraction results to the load prediction values ​​for consecutive control periods. Compare the predicted load values ​​with the training labels, and adjust the parameters of the temporal feature extraction layer and the prediction output layer according to the prediction deviation until the prediction deviation meets the convergence condition. Within the current control cycle, the latest continuous load data, corresponding time period characteristics, and ambient temperature are input into the trained load prediction model to generate an optimized load prediction sequence in the time domain. When the prediction deviation within a continuous control cycle meets the update conditions, the newly added actual load and corresponding training input will form an update sample to update the parameters of the load prediction model.

4. The adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in claim 3, characterized in that: The process of obtaining the state of charge and online available capacity includes accumulating the product of the battery current and the sampling time to update the state of charge, and correcting the state of charge according to the deviation between the measured value of the battery terminal voltage and the predicted value of the battery terminal voltage. By combining the relationship between battery voltage, battery current, and open-circuit voltage and state of charge, the battery internal resistance and online available capacity are recursively calculated, and the reliability of the online capacity estimation is characterized by the degree of fluctuation in voltage prediction deviation. The cumulative charge and discharge of the battery are used to calculate the long-term capacity decay based on the relationship between the cumulative result and the capacity decay corresponding to the battery temperature. The long-term capacity decay is then subtracted from the nominal battery capacity to obtain the long-term usable capacity. Based on the relative magnitudes of the reliability of online capacity estimation and the reliability of long-term capacity estimation, the fusion ratio of online available capacity and long-term available capacity is adjusted to obtain the fused available capacity; The health status of the fusion is characterized by the ratio of the fusion available capacity to the nominal capacity of the battery.

5. The adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in claim 4, characterized in that: The process of forming the comprehensive risk index includes generating battery heat according to the heat conversion relationship between battery current and battery internal resistance. Heat dissipation is generated by combining the temperature difference between the battery and the ambient temperature with wind speed, and radiant heat is generated according to the light intensity. The battery temperature is recursively calculated based on the heat balance relationship between the battery's heat generation, heat dissipation, and radiant heat. Then, the deviation between the recursively calculated temperature and the collected battery temperature is corrected to obtain the internal temperature of the battery. Temperature risk values ​​are formed based on the degree to which the internal temperature of the battery deviates from the safe temperature range, humidity risk values ​​are formed based on the degree to which the ambient humidity deviates from the safe humidity range, and health risk values ​​are formed based on the degree to which the integrated health status decreases relative to the health status baseline. The temperature risk value, humidity risk value, and health risk value are normalized and weighted to obtain a comprehensive risk index.

6. The adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in claim 5, characterized in that: The process of forming a dynamic safe operating boundary includes establishing a mapping relationship between the integrated health status, battery internal temperature, and basic charge and discharge capacity. Basic charge and discharge capabilities include basic charging power limits and basic discharging power limits; Substitute the current fusion health status and the current internal battery temperature into the mapping relationship, and obtain the basic charging power limit and the basic discharging power limit according to the charging and discharging capacity change relationship in the corresponding interval; The basic charging power limit and the basic discharging power limit are reduced in tandem with the increase in the comprehensive risk index to obtain the safe charging power limit and the safe discharging power limit. The lower limit of the basic state of charge and the upper limit of the basic state of charge are increased and decreased according to the increase of the comprehensive risk index to obtain the lower limit of the safe state of charge and the upper limit of the safe state of charge. The safe charging power limit, safe discharging power limit, safe state of charge limit, and safe state of charge upper limit are combined to form a dynamic safe operating boundary.

7. The adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in claim 6, characterized in that: The rolling optimization under dynamic safe operation boundary constraints includes dividing the optimization time domain into continuous control periods and setting the charging and discharging power of each control period as the power to be optimized. The power grid interaction power for each control period is generated based on the power balance relationship between the load forecast sequence and the power to be optimized, and the corresponding electricity price data for each control period is read. The grid energy consumption cost is formed by combining grid interaction power and electricity price data; the battery aging cost is formed by combining the power to be optimized, the integrated health status, and the battery internal temperature; and the operation risk cost is formed by combining the power to be optimized and the comprehensive risk index. The grid energy consumption cost, battery aging cost, and operational risk cost are accumulated for each control period to form an optimization target; The power to be optimized is constrained by safe charging power limit and safe discharging power limit. The state of charge is recursively calculated according to the battery energy change caused by the power to be optimized, and the state of charge is restricted between the lower limit of safe state of charge and the upper limit of safe state of charge. Under the condition that the power exchange between the grids meets the power exchange capacity range, obtain the charging and discharging power sequence with the lowest optimization objective; Execute the first charge / discharge power command corresponding to the current control period in the charge / discharge power sequence, and perform optimization again based on the updated multi-source operating data sequence after the current control period ends.

8. An adaptive optimization system for charging and discharging strategies for energy storage cabinets, employing the adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in any one of claims 1 to 7, characterized in that: It includes a multi-source data module, a load forecasting module, a state fusion module, a risk assessment module, a safety boundary module, and a rolling optimization module; The multi-source data module is used to collect load data, environmental data, and battery operation data, and form a multi-source operation data sequence according to the control cycle. The load forecasting module is used to establish a load forecasting model based on multi-source operational data sequences and generate load forecasting sequences. The state fusion module is used to estimate the state of battery operation data, obtain the state of charge and online available capacity, and fuse them with the long-term available capacity to form a fused health state. The risk assessment module is used to estimate the internal temperature of the battery by combining battery operation data, environmental data, and integrated health status, and form a comprehensive risk index. The safety boundary module is used to map the fusion health status and battery internal temperature to the basic charge and discharge capacity, and tighten the range of charge and discharge power and state of charge according to the comprehensive risk index to form a dynamic safety operation boundary. The rolling optimization module is used to perform rolling optimization under dynamic safe operation boundary constraints, output the first charge / discharge power command, and execute it cyclically as the data is updated.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the adaptive optimization method for charging and discharging strategies for energy storage cabinets as described in any one of claims 1 to 7.

Images