CN-122020537-A - Cross-scale joint estimation method and system for dynamic power scheduling of large energy storage system

CN122020537ACN 122020537 ACN122020537 ACN 122020537ACN-122020537-A

Abstract

The invention discloses a cross-scale joint estimation method for dynamic power scheduling of a large energy storage system, which comprises the steps of obtaining operation data in the energy storage system, constructing a corresponding first data set, constructing a first regression model for predicting SOC and a second regression model for predicting SOH based on an extreme gradient lifting tree of Bayesian optimization, forming a second data set by working condition data, a corresponding predicted SOC sequence and a predicted SOH sequence, constructing a bidirectional interaction neural network, comprising a main feature coding module and a coupled interaction modeling module, deploying the prediction model at an operation control end of the energy storage system, and predicting current SOH and SOC according to the collected working condition data. The invention also provides a cross-scale joint estimation system. The method provided by the invention can realize high-precision combined prediction of the battery state and enhance the safety, precision and robustness of the power scheduling of the energy storage system.

Inventors

Zheng Yelai
ZHANG XIN
LIU XUEQI
CAO HONGFU
BAO XIAOKUN

Assignees

上海交通大学

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. The cross-scale joint estimation method for dynamic power scheduling of the large energy storage system is characterized by comprising the following steps of: acquiring operation data in an energy storage system, wherein the operation data comprises working condition data, a historical SOC sequence and a historical SOH sequence, rearranging the working condition data in a charge-discharge cycle sequence, and constructing a corresponding first data set with the historical SOC sequence and the historical SOH sequence; Constructing a first regression model for predicting the SOC and a second regression model for predicting the SOH based on the Bayesian optimized extreme gradient lifting tree; respectively inputting the working condition data into a first regression model and a second regression model to output a predicted SOC sequence and a predicted SOH sequence, and forming a second data set by the working condition data, the corresponding predicted SOC sequence and the predicted SOH sequence; constructing a bidirectional interaction neural network, which comprises a main feature coding module and a coupling interaction modeling module; The main feature coding module is used for carrying out multi-mode feature extraction on input operation data so as to generate a joint embedded representation; The coupling interaction modeling module comprises an SOC leading path unit, an SOH leading path unit and a data enhancing unit, wherein the SOC leading path unit takes a historical SOC sequence and charge-discharge behaviors as input to predict an SOH result, the SOH leading path unit takes the historical SOH sequence and working condition data, the SOH leading path unit predicts the SOC result in a mode of pushing back the SOH to be the SOC, and the data enhancing unit is used for taking the output of one leading path unit as the self-attention input of the other leading path unit and carrying out condition adjustment on the other leading path unit; Performing primary training on the bidirectional interaction neural network by using the second data set, and performing secondary training on the bidirectional interaction neural network after the primary training by using the first data set so as to obtain a prediction model for predicting SOH and SOC; And deploying the prediction model at an operation control end of the energy storage system, and predicting the current SOH and SOC according to the collected working condition data.
2. The method of claim 1, wherein the operating condition data includes voltage, current, temperature, charge and discharge amount and number of cycles.
3. The cross-scale joint estimation method for dynamic power scheduling of a large energy storage system according to claim 1, wherein the bidirectional interaction neural network is constructed based on a collaborative transducer structure, a dynamic feedback relation between the bidirectional interaction neural network and the collaborative transducer structure is simulated through a residual coupling block, and nonlinear mapping between charge-capacity and impedance-health in a battery degradation evolution process is fused.
4. The method for cross-scale joint estimation of dynamic power scheduling of a large energy storage system according to claim 1, wherein in the second training process, a joint loss function is used to fine-tune the initial model.
5. The method of cross-scale joint estimation for dynamic power scheduling of large energy storage systems of claim 4, wherein the expression of the joint loss function is as follows: Wherein, the method comprises the steps of, In order to account for the overall loss, And For the weight coefficient set for the person, Weak tags output for a first regression model based on predicted SOC and a second regression model for predicting SOH 、 Loss in the direction of the flow, For losses under the guidance of real SOC and SOH tags, For physical coupling consistency loss, it is constructed from the known thermodynamic and electrochemical relationships of the cell.
6. The method of cross-scale joint estimation for dynamic power scheduling of large energy storage systems of claim 1, wherein the expression of the SOC dominant path element is as follows: Wherein, the method comprises the steps of, Charge and discharge charge of the period; is the depth of discharge for that period, k 1 , k 2 , And Is a preset experience parameter.
7. The method of cross-scale joint estimation for dynamic power scheduling of large energy storage systems of claim 1, wherein the SOH dominant path element is expressed as follows: ; where I represents the current signal and C nominal represents the nominal battery capacity.
8. The method of claim 1, wherein the predictive model further comprises an uncertainty estimate based on Monte Carlo Dropout methods before outputting the predicted result.
9. The cross-scale joint estimation method for dynamic power scheduling of a large energy storage system according to claim 1, wherein the parameters of the prediction model deployed in the energy storage system are iteratively updated through a lightweight error back propagation mechanism and a gradient constraint updating strategy.
10. A cross-scale joint estimation system, characterized by the steps for performing a cross-scale SOC and SOH joint estimation method for dynamic power scheduling of a large energy storage system as claimed in any of claims 1-9.

Description

Cross-scale joint estimation method and system for dynamic power scheduling of large energy storage system Technical Field The invention relates to the technical field of lithium ion battery state estimation, in particular to a cross-scale joint estimation method and system for dynamic power scheduling of a large energy storage system. Background Along with the continuous rising of the new energy grid-connected proportion, the large-scale battery energy storage system has become an important support for peak regulation and frequency modulation, accident standby and new energy consumption of the power system. However, in the actual operation process, the energy storage system faces complex challenges such as frequent charge-discharge switching, high-rate load change, long-term aging degradation and the like, and higher requirements are put on the safety, the accuracy and the dynamic response capability of power scheduling. The power scheduling process of the energy storage system is highly dependent on accurate perception of the battery State, wherein the State of Charge (SOC) reflects the current remaining available energy, and the State of Health (SOH) describes the degree of performance degradation of the battery capacity decay, internal resistance increase, etc. The combined estimation of the two is of great significance in avoiding battery overcharge and over-discharge, improving power utilization efficiency and realizing life-sensing type scheduling. However, the current mainstream scheduling methods generally suffer from the following disadvantages: Based on SOC scheduling only, SOH influence is ignored, so that the power capability cannot be estimated correctly after the battery is aged, and overload or mismatch is easy to cause; SOC and SOH estimation are usually carried out separately, the coupling relation and time scale difference between the SOC and the SOH are not considered, and unified modeling and high-precision prediction are difficult to realize; in a system with severe working condition changes and obvious inter-cluster differences, the traditional static model is difficult to effectively reflect state non-uniformity of cross-cell and cross-module, and the scheduling strategy lacks flexibility and robustness. Patent CN 119355530A discloses a method and a system for jointly estimating SOC-SOH based on biLSTM, wherein the method firstly acquires time sequence data of voltage, temperature, current, SOC and SOH in the current cycle period of an energy storage device, inputs the time sequence data of the voltage and the temperature into a temperature and voltage prediction sub-network, outputs a voltage prediction curve and a temperature prediction curve of the next cycle period, inputs the time sequence data of the voltage prediction curve, the temperature prediction curve and the current into the SOC prediction sub-network, outputs an SOH prediction curve, and inputs the time sequence data of the voltage prediction curve, the temperature prediction curve, the current and the SOH prediction curve into the SOH prediction sub-network, and outputs the SOC prediction curve. However, the method only agrees with the sequence of SOH and SOC prediction, does not perform cooperative optimization, cannot embody the relationship of the mutual coupling influence of SOH and SOC, and has limited prediction precision. The patent CN 120121995A discloses a method and a system for jointly estimating the SOC and the SOH of all the battery cells of an energy storage system, wherein the method firstly measures the terminal voltage of each battery, takes the battery cell with the largest terminal voltage and the battery cell with the smallest terminal voltage as characteristic battery cells, establishes a circuit model of the characteristic battery cells again, jointly estimates the SOC and the SOH of the characteristic battery cells based on the circuit model of the characteristic battery cells, jointly estimates the SOC and the SOH of each battery cell based on the circuit model of each battery cell, and corrects the SOC estimated value and the SOH estimated value of each battery cell obtained by the joint estimation by using the SOC estimated value and the SOH estimated value of each battery cell obtained by the joint estimation as a joint estimation result of the SOC and the SOH of all the battery cells of the energy storage system. However, as terminal voltage and circuit model parameters can change obviously along with the aging of the battery, the selection of the characteristic battery cell and the accuracy of the battery cell circuit model are affected, and therefore, the estimation accuracy of the method in the long-term operation of the energy storage system is limited. Disclosure of Invention The invention aims to provide a cross-scale joint estimation method and a system for dynamic power scheduling of a large energy storage system, which can realize high-precision joint prediction of battery states a