An energy storage system intelligent charging and discharging optimization method and system based on electricity price and capacity prediction

Abstract

A kind of energy storage system intelligent charge-discharge optimization method and system based on electricity price and capacity prediction.Electricity price prediction model, battery capacity attenuation prediction model, prediction-driven charge-discharge optimization module, control execution module and self-learning feedback module are included in the method and system.The method and system obtain multi-source operation data through data acquisition module and preprocess, predict future electricity price sequence using LSTM time series deep learning model, and predict battery state of health (SOH) by fusing "electrochemical mechanism model + LSTM correction model".The charge-discharge optimization module constructs multi-objective function based on the above prediction results, combines SOC, power, temperature and other constraint conditions, and generates optimal charge-discharge strategy using model predictive control.The method and system aim to solve the technical problems that existing energy storage charge-discharge strategy cannot perceive electricity price changes, does not consider battery capacity attenuation, relies on empirical rules and lacks closed-loop optimization.

Description

Technical Field

[0001] This invention relates to the field of electrical technology, specifically to the charging and discharging control technology of energy storage systems, and more particularly to an intelligent charging and discharging optimization method and system for energy storage systems based on electricity price and capacity prediction. Background Technology

[0002] With the rapid expansion of renewable energy grid connection, battery energy storage systems (BESS) play an irreplaceable and crucial role in scenarios such as peak shaving and valley filling, frequency and voltage regulation, and improving power quality. The operating efficiency, battery life, and operational stability of energy storage systems largely depend on the scientific and rational nature of their charging and discharging strategies.

[0003] However, existing energy storage systems generally suffer from the following technical shortcomings in their charging and discharging strategies: 1. The inability to foresee future electricity price trends leads to energy storage systems discharging during periods of non-optimal electricity prices, resulting in low energy utilization efficiency and failure to maximize economic benefits. 2. The degradation characteristics of battery capacity due to factors such as operating time and number of cycles were not fully considered, which may easily lead to over-cycle or deep discharge, accelerate battery aging, and shorten battery life. 3. Control strategies are mostly designed based on empirical rules, lacking the ability to predict future operating environments (such as load and weather changes), resulting in poor adaptability; 4. The lack of a prediction-driven closed-loop optimization mechanism makes it impossible to dynamically adjust the charging and discharging strategy based on future electricity price trends and battery capacity changes, resulting in difficulty in continuously optimizing the strategy execution effect.

[0004] Therefore, there is an urgent need for an energy storage system operation technology that can combine electricity price forecasting, battery capacity degradation forecasting, and control strategy optimization to improve the energy efficiency and battery life of the energy storage system and enhance operational stability through technical means. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent charging and discharging optimization method and system for energy storage systems based on electricity price and capacity prediction, achieving the following objectives: predicting future electricity price cycles and battery capacity degradation trends in advance; dynamically generating optimal charging and discharging strategies to maximize energy gains; reducing battery losses and extending battery lifespan; realizing intelligent and adaptive charging and discharging strategies to improve the operational stability of the energy storage system; and constructing a closed-loop optimization mechanism to continuously improve system performance.

[0006] A smart charging and discharging optimization method for energy storage systems based on electricity price and capacity prediction includes the following steps: S1: Data Acquisition and Preprocessing: Collect electricity price data, load data, meteorological data, and battery operation data, and sequentially perform time alignment, sliding window mean filling of missing values, Savitzky-Golay filtering for smoothing and noise reduction, Min-Max normalization, and sequence windowing to obtain standardized input data; S2: Electricity price forecast: Input the standardized data processed in step S1 into the LSTM network model, train it with the mean squared error loss function, and output the electricity price forecast sequence for the next 24 hours or 7 days. S3: Battery capacity degradation prediction: S31: Construct an electrochemical mechanism decay model, introduce a dynamic decay coefficient related to SOH, and the modified model expression is as follows: ; in: Indicates the initial health state of the battery; Depth of discharge; This represents the cumulative number of loops. Operating temperature; For reference temperature; and These represent the decay coefficients for cyclic aging and temperature aging, respectively. S32: Based on battery cycle life, depth of discharge, temperature, charge / discharge rate, and As input, the LSTM correction model outputs an error correction term. ; S33: Integrating the output of the electrochemical mechanism decay model with the error correction term, we obtain... ; S4: Charge / Discharge Optimization: Based on Electricity Price Prediction Sequence Based on parameters such as the current SOC state and the upper limit of charge / discharge power, a multi-objective optimization function is constructed: ; The battery degradation cost item is defined as follows: ; in The penalty coefficient is... The health perception weighting function based on SOH prediction is defined as follows: ; in, For weight adjustment parameters; Combining SOC range constraints, charge / discharge power constraints, temperature constraints, and depth of discharge limitations, the optimal charge / discharge strategy is obtained by using the MPC framework for rolling solution. S5: Control Execution and Self-Learning Feedback: The optimal charging and discharging strategy is sent to the energy storage device for execution. Incremental model training is triggered every 7 days. The model weights and optimizer parameters are updated based on actual operating data. Rapid retraining is triggered under extreme conditions.

[0007] A smart charging and discharging optimization system for energy storage systems based on electricity price and capacity prediction, comprising: The data acquisition module is used to collect electricity price data, load data, meteorological data, and battery operation data, and to perform time alignment, missing value imputation, smoothing and denoising, normalization, and feature engineering processing on the data. The electricity price prediction model uses a time-series deep learning LSTM network as input, and outputs a predicted electricity price sequence for the next 24 hours or 7 days. The battery capacity decay prediction model adopts a fusion structure of "electrochemical mechanism model + LSTM correction model". The inputs are battery cycle number, depth of discharge, operating temperature, charge and discharge rate and historical health status, and the output is the predicted value of future battery health status SOH(t). The prediction-driven charge-discharge optimization module constructs a multi-objective function with the goal of "maximizing energy gain and minimizing capacity loss" based on the electricity price prediction sequence and SOH(t). Combining the SOC range, maximum charge-discharge power, battery temperature, and cycle number constraints, the module uses model predictive control to generate the optimal charge-discharge strategy. The strategy includes the charging power, discharging power, SOC target range, discharge depth limit, and cycle number limit for each time period. The control execution module communicates with EMS / PCS / BMS to issue the optimal charging and discharging strategy, thereby realizing real-time adaptive control of the energy storage device. The self-learning feedback module records actual operating data and, based on prediction deviations and strategy execution effects, regularly updates and incrementally trains the parameters of the electricity price prediction model, the battery capacity degradation prediction model, and the charge / discharge optimization module.

[0008] Compared with the prior art, the present invention has the following beneficial effects: 1. Significantly improved energy utilization: By using the LSTM model to predict high and low electricity price cycles in advance, the energy storage system can discharge during the period with the optimal electricity price, improving energy utilization by 10-30% compared to the traditional empirical strategy, thus maximizing economic benefits. 2. Extended battery life: The battery capacity degradation is accurately predicted by the "mechanism + data" fusion model. Combined with the degradation penalty term in the multi-objective optimization function, deep discharge and excessive cycling are limited, reducing annualized capacity loss by 10-20% and effectively extending battery life. 3. High level of strategy intelligence: It eliminates the reliance on human experience and realizes real-time dynamic optimization of charging and discharging strategies based on predictive models, which can adapt to changes in electricity price, load, weather and battery status. 4. Enhanced operational stability: Through multiple safeguards such as temperature constraints, SOC range constraints, and DOD limits, risks such as abnormal battery temperature and excessive deep discharge are avoided, thereby improving the operational safety and stability of the energy storage system. 5. Closed-loop optimization and continuous upgrade: Through the self-learning feedback module, the model and optimization parameters are continuously updated using actual operating data, so that good operating performance can be maintained even under extreme conditions, and long-term optimization of system performance can be achieved. Attached Figure Description

[0009] Figure 1 This is the overall architecture diagram of the intelligent charging and discharging optimization system for the energy storage system of the present invention.

[0010] Figure 2 This is a schematic diagram of the electricity price prediction model of the present invention.

[0011] Figure 3 This is a schematic diagram of the battery capacity decay prediction model of the present invention. Detailed Implementation

[0012] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] like Figure 1 As shown, an intelligent charge-discharge optimization system for an energy storage system based on electricity price and capacity prediction is disclosed. The data acquisition / preprocessing module outputs data to the electricity price prediction model and the battery capacity decay prediction model. The outputs of the two prediction models are input to the prediction-driven charge-discharge optimization module. The charge-discharge optimization module outputs instructions to the control execution module. The control execution module is connected to the energy storage device. The operation data feedback unit collects the actual operation data of the device and inputs it to the self-learning feedback module. The self-learning feedback module establishes feedback connections with the electricity price prediction model, the battery capacity decay prediction model, and the charge-discharge optimization module, respectively.

[0014] It includes a data acquisition module, an electricity price prediction model, a battery capacity degradation prediction model, a prediction-driven charge and discharge optimization module, a control execution module, and a self-learning feedback module. The functions of each module are as follows: Data acquisition module: It is responsible for collecting multi-source operational data, including electricity price data, load data, meteorological data (temperature, irradiance, cloud cover, wind speed, etc.) and battery operation data (SOC, cycle count, charge / discharge rate, battery temperature, etc.). The collected raw data is preprocessed by performing time alignment (using the hour as a unified benchmark, high-frequency data aggregation, and low-frequency data forward filling), missing value imputation (3-window moving average method), smoothing and denoising (first-order Savitzky-Golay filter), normalization (Min-Max normalization), and feature engineering (sequence windowing, input sequence length 168) to form standardized input data.

[0015] Electricity price forecasting model: A temporal deep learning LSTM network is used as input, along with preprocessed historical electricity price, weather, and load data. The model training parameters were set as follows: the loss function was mean squared error (MSE), the batch size was 32, the number of training epochs was 100, overfitting was prevented by EarlyStopping (patience=10), and the learning rate was dynamically reduced by ReduceLROnPlateau (factor=0.5). Output the electricity price forecast sequence for the next 24 hours or 7 days to provide a basis for electricity price optimization of charging and discharging.

[0016] Battery capacity degradation prediction model: The fusion structure of "electrochemical mechanism model + LSTM correction model" is adopted to balance physical interpretability and prediction accuracy. Electrochemical mechanism degradation model: Introducing a dynamic degradation coefficient related to SOH corrects the defect of a fixed degradation coefficient in traditional models, enabling the characterization of nonlinear accelerated degradation characteristics in the later stages of battery life. The model expression is as follows: ; in: Indicates the initial health state of the battery; Depth of discharge; This represents the cumulative number of loops. Operating temperature; For reference temperature; and These represent the decay coefficients for cyclic aging and temperature aging, respectively.

[0017] LSTM calibration model: Inputs include battery cycle count, DOD, temperature, charge / discharge rate, and electrochemical mechanism model output. The network structure is LSTM (units=32) + LSTM (units=64) + Dense (units=1), and the output error correction term is... ; Fusion Output: This yields a predicted value for the future health status of the battery.

[0018] Predictive-driven charge / discharge optimization module (core module): Input parameters: electricity price forecast series, Current SOC status, maximum charge / discharge power, real-time temperature, and charge / discharge efficiency parameters; Construct a multi-objective optimization function: With the objective of "maximizing energy gain and minimizing capacity loss," energy gain = electricity price difference × discharge amount, and capacity loss is quantified through the attenuation cost term. The function expression is as follows:

[0019] The battery degradation cost item is defined as follows: ; in The penalty coefficient is... The health perception weighting function based on SOH prediction is defined as follows:

[0020] in, This is the weight adjustment parameter.

[0021] Constraints: (1) SOC range constraints: ;

[0022] in: ; (2) Charging and discharging power constraints: ; (3) Temperature constraint: When T>45℃, forced ; (4) DOD restrictions: ; .

[0023] Solution method: The MPC framework is used for rolling solution. The optimal charging and discharging strategy is obtained through preprocessing + convex optimization (QP solver such as OSQP, SCS) or mixed integer programming (MIQP solver such as Gurobi, CPLEX). The output is the charging power, discharging power, SOC target range, DOD limit and cycle number limit for each time period.

[0024] Control execution module: Establish communication with EMS (Energy Management System), PCS (Converter System), and BMS (Battery Management System). Communication protocols include Modbus / TCP (with PCS), CAN / RS485 (with BMS), and IEC 61850 (with Measurement and Control System). The optimal strategy output by the charge / discharge optimization module is sent to each device to execute charge / discharge power limits, SOC target control, DOD limit constraints, and charge / discharge allow / discharge commands, thereby achieving real-time adaptive control of the energy storage device.

[0025] Self-learning feedback module: Record actual operating data of the energy storage system (including actual electricity price, actual SOH, actual charging and discharging power, SOC changes, etc.) and store them in the data lake; trigger incremental model training every 7 days and update the weights of the electricity price prediction model and the capacity decay prediction model using prediction bias (electricity price prediction bias, SOH prediction bias); The optimizer parameters of the charge and discharge optimization module are updated based on the strategy execution results (actual values ​​of energy efficiency and capacity loss). When special operating conditions such as extreme weather or abnormal electricity prices occur, rapid retraining is triggered to ensure the adaptability of the model and strategy.

[0026] A smart charging and discharging optimization method for energy storage systems based on electricity price and capacity prediction includes the following steps: S1: Data Acquisition and Preprocessing: Collect multi-source running data, and obtain standardized input data through time alignment, missing value imputation, smoothing and denoising, normalization and sequence windowing. S2: Electricity Price Forecasting: Input standardized data into the LSTM network model for training, and output a future electricity price forecast sequence; S3: Battery capacity degradation prediction: This is achieved by fusing an "electrochemical mechanism model" with an "LSTM correction model," resulting in... ; S4: Charge / Discharge Optimization: Based on Electricity Price Prediction Sequence and A multi-objective optimization function is constructed, and the optimal charging and discharging strategy is obtained by solving the constraint conditions through MPC. S5: Control Execution and Self-Learning Feedback: Issue and execute the optimal strategy, and continuously update the model and optimizer parameters based on actual operating data.

[0027] Example 1: Implementation of the Electricity Price Forecasting Model This embodiment provides a technical solution for predicting future electricity market prices, used to achieve predictive-driven dispatching of energy storage systems. Combined with... Figure 2 The raw data is processed (time alignment, missing value imputation, etc.) and then input into the LSTM network. After training, it outputs an electricity price prediction sequence.

[0028] 1.1 Data Processing The data sources for the electricity price forecasting module include: electricity price data, load data, meteorological data (temperature, irradiance, cloud cover, wind speed, etc.), and battery operation data (SOC, cycle count, capacity, temperature). All of the above raw multi-source data are collectively denoted as f0 = {raw_price, raw_load, raw_weather, raw_battery}.

[0029] Because the above data come from different sources, their time granularity, timestamp alignment, and data integrity vary. Therefore, they need to be processed in the following order to form standardized input data that can be used by the prediction model.

[0030] ① Time Alignment: Using the hour as a unified time base, time alignment is performed on all types of data in f0. High-frequency data is aggregated hourly, while low-frequency data is padded forward to ensure all features correspond on the same time axis. After time alignment, a unified multivariate time series data f1 is obtained, in the following form: f1[t] = {timestamp_t, price_t, load_t, weather_t, battery_t}.

[0031] The f1 is used as the basic input for subsequent data cleaning and feature processing.

[0032] ② Missing Value Handling: For missing values ​​in f1 caused by abnormal acquisition or communication interruption, a sliding window mean method is used for imputation. The sliding window length is set to k = 3, and the mean of adjacent time points is used to replace the missing values, thereby ensuring the continuity of the time series. After missing value handling, the output data is denoted as f2, which serves as the input data for the smoothing step.

[0033] ③ Smoothing: To eliminate abnormal spike noise in continuous variables such as electricity price and load, a first-order Savitzky-Golay filter is applied to the key time series features in f2 for smoothing while maintaining the overall trend, resulting in noise-suppressed time series data f3. f3 is used for subsequent feature scaling.

[0034] ④ Feature Normalization: To address the issue of significant differences in the dimensions of different features in f3, Min-Max normalization is applied to each feature dimension to improve the numerical stability and convergence speed of the neural network training. The normalization parameters are fitted to the training dataset and maintained consistently during the prediction phase. The normalized output data is denoted as f4.

[0035] ⑤ Sequence Windowing: The continuous time series data f4 is converted into supervised learning samples. The input sequence length is set to past_steps = 168 (corresponding to 7 consecutive days of historical data), and a sliding window method is used to construct the sample data. Each input sample consists of the feature sequence of the past 168 time steps, and the prediction target is the electricity price at the current moment. After sequence windowing, the model input and output dataset f5 is obtained, which includes the input feature matrix X and the corresponding label y.

[0036] Finally, f5 was used as the input data for the electricity price forecasting model.

[0037] 1.2 Model Structure A multi-layer LSTM network is used, with the following structure: hierarchy type parameter Layer 1 LSTM units=64, return_sequences=True Layer 2 LSTM units=128, return_sequences=True Layer 3 LSTM units=64, return_sequences=False Dense Fully connected layer units=24, outputs the electricity price for the next 24 hours. Activation function tanh Optimizer Adam (learning_rate=0.001) 1.3 Training Methods Loss function: Mean Squared Error (MSE); Batch size: 32; Number of training rounds: 100; Use EarlyStopping (patience=10) to prevent overfitting; Use ReduceLROnPlateau (factor=0.5) to dynamically reduce the learning rate.

[0038] 1.4 Output The trained electricity price prediction model outputs a 24-hour electricity price prediction sequence, denoted as: ;

[0039] The prediction results This serves as one of the input data for subsequent energy storage optimization and scheduling algorithms.

[0040] Example 2: Implementation of the capacity decay prediction model This embodiment provides a technical model for predicting the degradation trend of battery capacity over time. By predicting the battery's health status, it dynamically limits the operating depth of the energy storage system, thereby extending battery life. Combined with... Figure 3 The battery capacity decay prediction model of the present invention uses running data (number of cycles, DOD, temperature T, historical SOH, C-rate) which, after feature processing (serialization, smoothing and denoising, and anomaly filtering), are input into the mechanism model and the LSTM correction model respectively. The predicted SOH is output after the two models are fused.

[0041] 2.1 Data Input The data inputs for the capacity degradation prediction model include: battery cycle count, depth of discharge (DOD) for each cycle, battery temperature, charge / discharge rate, and historical state of health (SOH). All of the above data are used as the raw inputs to the capacity degradation prediction model.

[0042] 2.2 Feature Processing The input data is processed as follows: First, the features based on the number of iterations are converted into time series format; second, each feature is smoothed using a 7-day moving average method; simultaneously, abnormal temperature data is filtered out, and data with temperatures above 60℃ are removed. After feature processing, continuous and stable time series feature data are formed for model calculation.

[0043] 2.3 Model Structure A hybrid approach combining "mechanism model + data-driven model" is adopted: ① Electrochemical mechanism decay model (basic) ; in: Indicates the initial health state of the battery; Depth of discharge; This represents the cumulative number of loops. Operating temperature; For reference temperature; and These represent the decay coefficients for cyclic aging and temperature aging, respectively.

[0044] This model possesses good physical interpretability and can reflect the basic degradation trend of batteries under different operating conditions. However, its default assumption that the degradation coefficient remains constant throughout the entire lifespan makes it difficult to accurately describe the nonlinear accelerated degradation phenomenon caused by internal mechanism changes in the later stages of the battery's lifespan. To overcome the limitation of the fixed degradation coefficient in traditional mechanistic models, this invention introduces a state-aware degradation coefficient correction mechanism, enabling the cyclic aging and temperature aging rates to dynamically change with the battery's health state. Specifically, the constant degradation coefficient in the original model is corrected to a dynamic coefficient related to SOH (State of Health):

[0045] Among them, the correction function Defined as:

[0046] In the formula, Indicates the attenuation amplification factor. This represents the nonlinear enhancement index, used to adjust the degree of acceleration in decay.

[0047] The correction mechanism has the following characteristics: when the battery is in a healthy state (SOH≈1), the correction coefficient approaches 1, thus maintaining the physical consistency of the original mechanism model in the early stage of the lifespan; as SOH decreases, the correction term gradually amplifies the decay rate, effectively characterizing the nonlinear accelerated aging characteristics in the later stage of the lifespan caused by mechanisms such as SEI film thickening and electrode structure degradation.

[0048] After introducing state-aware correction, the electrochemical mechanism decay model can be expressed as: ; ②LSTM calibration model Input: [cycle_count, DOD, T, C_rate, ]; Structure: LSTM (units=32), LSTM (units=64), Dense (units=1); Output: Error correction term for the mechanistic model .

[0049] ③ Fusion Output ; Finally, the output of the mechanistic model is added to the LSTM correction result to obtain the predicted value of the battery health state. .

[0050] Example 3: Implementation process of optimization algorithms (MPC / DP / RL) This embodiment provides a technical solution for generating the optimal charging and discharging strategy based on prediction results.

[0051] 3.1 Input Data Prediction results from the first two examples: ① Future electricity price sequence ; ② Future capacity decay ; ③ Current SOC status Upper limit of charging and discharging power Real-time temperature Charge and discharge efficiency parameters .

[0052] 3.2 Optimize the model structure (1) Optimize the objective function Building upon the traditional profit maximization objective, this paper introduces a degradation penalty term related to battery health state to construct a health-aware multi-objective optimization function:

[0053] The battery degradation cost item is defined as follows: ; in The penalty coefficient is... The health perception weighting function based on SOH prediction is defined as follows:

[0054] in, This is the weight adjustment parameter.

[0055] (2) State transition and energy conservation constraints The dynamic evolution of battery SOC satisfies the following energy conservation relationship: ; (3) SOC range: ; in: ; (5) Charge and discharge power constraints: ; (6) Temperature constraint: When T>45°C, forced ; Avoid excessively deep loops: ; in: .

[0056] 3.3 Solver This invention employs a model predictive control (MPC) framework for rolling solution. Considering that the objective function contains quadratic terms and the constraints include logical states, one of the following two solution methods can be used: 1. Preprocessing + Convex Optimization Approach: In each control step, the boundary conditions of the decision variables are preprocessed based on the predicted temperature sequence and SOH state (e.g., if the predicted temperature T>45 degrees Celsius for a certain period, the power upper limit for that period is directly set to 0). Subsequently, the problem is transformed into a standard quadratic programming (QP) problem and solved quickly using solvers such as OSQP or SCS.

[0057] 2. Mixed Integer Programming Approach: Introducing binary variables representing the charging and discharging states, the problem is modeled as a mixed integer quadratic programming (MIQP) approach. Solvers such as Gurobi, CPLEX, or CBC are used to accurately handle the charging and discharging mutual exclusion constraints.

[0058] Finally, the optimal control commands for the next scheduling cycle (e.g., 15 minutes) are obtained: charging power, discharging power, target SOC, and dynamic DOD limit.

[0059] Example 4: Strategy Execution and Feedback Self-Learning Module This embodiment realizes the execution, correction, and self-learning capabilities of the aforementioned optimization strategy.

[0060] 4.1 Control Execution Process The policies output by MPC are distributed through the EMS system: Communication with PCS (converter): Modbus / TCP; Communication with BMS: CAN / RS485; Communication with the measurement and control system: IEC 61850; The execution content includes: charge and discharge power limits, SOC targets, DOD limits, and instructions to prohibit or permit charge and discharge.

[0061] 4.2 Online self-learning The system continuously improves performance through the following steps: ① Store actual operational data in a data lake ② Trigger incremental model training once every 7 days ③ Prediction bias is used to update model weights ④ The strategy execution results are used to update optimizer parameters. ⑤ Trigger rapid retraining when extreme weather or abnormal electricity prices occur.

[0062] Those skilled in the art should understand that the above embodiments are merely specific implementations of the present invention and are not intended to limit the scope of protection of the present invention. Any equivalent substitutions or modifications made based on the technical solutions of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A smart charging and discharging optimization method for energy storage systems based on electricity price and capacity prediction, characterized in that, Includes the following steps: S1: Data Acquisition and Preprocessing: Collect electricity price data, load data, meteorological data, and battery operation data, and sequentially perform time alignment, sliding window mean filling of missing values, Savitzky-Golay filtering for smoothing and noise reduction, Min-Max normalization, and sequence windowing to obtain standardized input data; S2: Electricity price forecast: Input the standardized data processed in step S1 into the LSTM network model, train it with the mean squared error loss function, and output the electricity price forecast sequence for the next 24 hours or 7 days. S3: Battery capacity degradation prediction: S31: Construct an electrochemical mechanism decay model, introduce a dynamic decay coefficient related to SOH, and the modified model expression is as follows: ； in: Indicates the initial health state of the battery; Depth of discharge; This represents the cumulative number of loops. Operating temperature; For reference temperature; and These represent the decay coefficients for cyclic aging and temperature aging, respectively. S32: Based on battery cycle life, depth of discharge, temperature, charge / discharge rate, and As input, the LSTM correction model outputs an error correction term. ; S33: Integrating the output of the electrochemical mechanism decay model with the error correction term, we obtain... ; S4: Charge / Discharge Optimization: Based on Electricity Price Prediction Sequence Given the current SOC state and the upper limit of charge / discharge power, a multi-objective optimization function is constructed: ； The battery degradation cost item is defined as follows: ； in The penalty coefficient is... The health perception weighting function based on SOH prediction is defined as follows: ； in, For weight adjustment parameters; Combining SOC range constraints, charge / discharge power constraints, temperature constraints, and depth of discharge limitations, the optimal charge / discharge strategy is obtained by using the MPC framework for rolling solution. S5: Control Execution and Self-Learning Feedback: The optimal charging and discharging strategy is sent to the energy storage device for execution. Incremental model training is triggered every 7 days. The model weights and optimizer parameters are updated based on actual operating data. Rapid retraining is triggered under extreme conditions.

2. The optimization method according to claim 1, characterized in that, The input sequence length for sequence windowing processing in step S1 is 168, corresponding to 7 consecutive days of historical data.

3. The optimization method according to claim 1, characterized in that, The temperature constraint mentioned in step S4 is: when the battery temperature T>45℃, forced .

4. The optimization method according to claim 1, characterized in that, The depth of discharge limit mentioned in step S4 is .

5. The optimization method according to claim 1, characterized in that, In step S4, the solution method for the MPC framework is quadratic programming or mixed integer quadratic programming, and the corresponding solvers include OSQP, SCS, Gurobi, CPLEX or CBC.

6. A smart charging and discharging optimization system for an energy storage system based on electricity price and capacity prediction, used to implement the optimization method described in any one of claims 1-5, characterized in that, include: The data acquisition module is used to collect electricity price data, load data, meteorological data, and battery operation data, and to perform time alignment, missing value imputation, smoothing and denoising, normalization, and feature engineering processing on the data. The electricity price prediction model uses a time-series deep learning LSTM network as input, and outputs a predicted electricity price sequence for the next 24 hours or 7 days. The battery capacity decay prediction model adopts a fusion structure of "electrochemical mechanism model + LSTM correction model". The inputs are battery cycle number, discharge depth, operating temperature, charge and discharge rate and historical health status, and the output is the predicted value of future battery health status SOH(t). The prediction-driven charge-discharge optimization module constructs a multi-objective function with the goal of "maximizing energy gain and minimizing capacity loss" based on the electricity price prediction sequence and SOH(t). Combining the SOC range, maximum charge-discharge power, battery temperature, and cycle number constraints, the module uses model predictive control to generate the optimal charge-discharge strategy. The strategy includes the charging power, discharging power, SOC target range, discharge depth limit, and cycle number limit for each time period. The control execution module communicates with EMS / PCS / BMS to issue the optimal charging and discharging strategy, thereby realizing real-time adaptive control of the energy storage device. The self-learning feedback module records actual operating data and, based on prediction deviations and strategy execution effects, regularly updates and incrementally trains the parameters of the electricity price prediction model, the battery capacity degradation prediction model, and the charge / discharge optimization module.

7. The optimization system according to claim 6, characterized in that, The meteorological data includes temperature, irradiance, cloud cover, and wind speed, while the battery operating data includes SOC, cycle count, charge / discharge rate, and battery temperature.

8. The optimization system according to claim 6, characterized in that, The training parameters for the electricity price prediction model are: batch size 32, number of training rounds 100, EarlyStopping to prevent overfitting, and ReduceLROnPlateau to dynamically adjust the learning rate.

9. The optimization system according to claim 6, characterized in that, The communication protocols between the control execution module and the EMS / PCS / BMS include Modbus / TCP, CAN / RS485 and IEC 61850.

Images