Search

CN-122017566-A - SOH prediction method and system for energy storage battery

CN122017566ACN 122017566 ACN122017566 ACN 122017566ACN-122017566-A

Abstract

The invention discloses an energy storage battery SOH prediction method and system, and belongs to the technical field of energy storage battery management. The method comprises the steps of collecting operation data of a plurality of groups of single battery cells in real time under a plurality of continuous time steps, preprocessing the operation data to generate a time-battery cell two-dimensional input matrix, inputting the time-battery cell two-dimensional input matrix into a pre-trained self-adaptive space-time fusion prediction model, outputting SOH predicted values of all the single battery cells, and evaluating the health state of the battery cells according to the SOH values.

Inventors

  • Ji Tengwei
  • XU DAYONG
  • SHEN YU

Assignees

  • 合肥国轩高科动力能源有限公司

Dates

Publication Date
20260512
Application Date
20260120

Claims (10)

  1. 1. A method for predicting SOH of an energy storage battery, the method comprising: Acquiring operation data of a plurality of groups of single battery cells in real time under a plurality of continuous time steps, and preprocessing the operation data to generate a time-battery cell two-dimensional input matrix; inputting a time-cell two-dimensional input matrix into a pre-trained self-adaptive space-time fusion prediction model, outputting SOH predicted values of all the single cells, and evaluating the health state of the cells according to the SOH values; wherein the adaptive spatiotemporal fusion prediction model comprises: The convolution neural network CNN is used for convolving each cell along the time dimension and extracting the local time sequence mutation characteristics of the single cell; ConvTransformer layers are used for capturing the dependency relationship between the battery cells and the long sequence rule according to the local time sequence mutation characteristics of the battery cells and outputting time-space fusion characteristics; Stage self-adaptive LSTM, according to the space-time fusion characteristic, carrying out cell dimension polymerization, inputting the polymerized time sequence into LSTM, and outputting SOH predicted value of each single cell.
  2. 2. The method for predicting the SOH of the energy storage battery according to claim 1, wherein the self-adaptive space-time fusion prediction model adopts a PSO-SA hybrid optimization module to combine a particle swarm optimization algorithm and a simulated annealing algorithm to optimize preset structural superparameters in the self-adaptive space-time fusion prediction model, so as to obtain a trained self-adaptive space-time fusion prediction model.
  3. 3. The method of claim 1, wherein the operational data includes a single level signature and a system level derived signature; the single-level characteristics comprise voltage, current, temperature and accumulated cycle times of each battery cell at each time step; the system level derivative features comprise voltage range, balanced current integral value, module temperature gradient and dynamic change rate of charge and discharge multiplying power of the battery pack under corresponding time steps.
  4. 4. The method of claim 1, wherein preprocessing the operation data to generate a time-cell two-dimensional input matrix comprises: The single-level features and the system-level derived features are aligned with the cell numbers according to time steps, and are organized into a space-time feature tensor taking time as a first dimension, the cell as a second dimension and the feature vector as a third dimension, and the space-time feature tensor is recorded as a time-cell two-dimensional input matrix.
  5. 5. The method of claim 1, wherein the preprocessing includes identifying and rejecting outlier data in the operating data using an orphan forest algorithm.
  6. 6. The method for SOH prediction of an energy storage battery according to claim 1, wherein the adaptive spatiotemporal fusion prediction model optimizes model hyper-parameters using a multi-objective fitness function, the fitness function being: ; Wherein, the The root mean square error is predicted for SOH, For the model to infer the time of day, In order to preset the maximum allowable inference time, Is a weight coefficient.
  7. 7. The method for predicting SOH of an energy storage battery according to claim 1, wherein the convolutional neural network CNN adopts 3-layer 1D-CNN to extract local abrupt patterns of single-cell timing characteristics.
  8. 8. The method of claim 1, wherein the ConvTransformer layers employ a sparse self-attention mechanism.
  9. 9. An energy storage battery SOH prediction system, the system comprising: The characteristic module is configured to acquire operation data of a plurality of groups of single battery cells in real time under a plurality of continuous time steps, and preprocesses the operation data to generate a time-battery cell two-dimensional input matrix; The self-adaptive space-time fusion model is configured to input a time-cell two-dimensional input matrix into a pre-trained self-adaptive space-time fusion prediction model, output SOH predicted values of all the single cells and evaluate the health state of the cells according to the SOH values.
  10. 10. The energy storage battery SOH prediction system of claim 9, wherein the adaptive spatiotemporal fusion prediction model comprises: The convolution neural network CNN is configured to convolve each cell along the time dimension and extract the local time sequence mutation characteristics of the single cell; ConvTransformer layers, which are configured to capture the dependency relationship and the long sequence rule of the cross-cell according to the local time sequence mutation characteristics of the cell and output the time-space fusion characteristics; And the stage self-adaptive LSTM is configured to perform cell dimension aggregation according to the space-time fusion characteristics, input the aggregated time sequence into the LSTM and output the SOH predicted value of each single cell.

Description

SOH prediction method and system for energy storage battery Technical Field The invention relates to an energy storage battery SOH prediction method and system, and belongs to the technical field of energy storage battery management. Background Lithium ion batteries have become a core component of energy storage systems due to their high energy density characteristics, wherein lithium iron phosphate batteries have taken up the mainstream market due to their long cycle life and high safety. The energy storage battery pack is generally formed by connecting hundreds of power cells in series and parallel, and under complex working conditions such as peak-valley arbitrage, standby power supply and the like, the problems of serious cell inconsistency, obvious aging rate difference, large temperature gradient change and the like exist, so that the traditional cell SOH prediction method is difficult to directly apply. In the prior art, three main limitations are that firstly, the cross-cell space-time correlation of an energy storage battery pack is not considered, key characteristics such as balanced current and voltage difference in the pack cannot be captured, secondly, a simple model has gradient attenuation problem in processing long-sequence and multi-node correlation analysis, thirdly, single PSO optimization is easy to sink into local optimum, and the decision of energy storage operation and maintenance is difficult to support. Therefore, the SOH prediction method which is adaptive to the characteristics of the energy storage battery and has high precision and strong interpretation is developed, and has important significance for improving the reliability and the economy of the energy storage system. Disclosure of Invention The invention aims to solve the problem of space-time characteristic modeling of a cross-cell of an energy storage battery pack in the prior art, provides an energy storage battery SOH prediction method and system, improves SOH prediction accuracy under complex working conditions, and meets SOH prediction requirements of cells under different energy storage scenes; in order to achieve the above purpose/solve the above technical problems, the present invention is realized by adopting the following technical scheme: an energy storage battery adaptive space-time fusion SOH prediction method, the method comprising: Acquiring operation data of a plurality of groups of single battery cells in real time under a plurality of continuous time steps, and preprocessing the operation data to generate a time-battery cell two-dimensional input matrix; inputting a time-cell two-dimensional input matrix into a pre-trained self-adaptive space-time fusion prediction model, outputting SOH predicted values of all the single cells, and evaluating the health state of the cells according to the SOH values; wherein the adaptive spatiotemporal fusion prediction model comprises: The convolution neural network CNN is used for convolving each cell along the time dimension and extracting the local time sequence mutation characteristics of the single cell; ConvTransformer layers are used for capturing the dependency relationship between the battery cells and the long sequence rule according to the local time sequence mutation characteristics of the battery cells and outputting time-space fusion characteristics; Stage self-adaptive LSTM, according to the space-time fusion characteristic, carrying out cell dimension polymerization, inputting the polymerized time sequence into LSTM, and outputting SOH predicted value of each single cell. According to the technical scheme, through the cooperation of the multiple modules, the prediction precision of the SOH value of the battery core is greatly improved, the inference time is optimized by means of the quantification of sparse attention and a model, the inference efficiency is improved, the real-time monitoring requirement of the energy storage BMS is met, the multi-type energy storage batteries such as lithium iron phosphate and ternary lithium are adapted, the requirement of accurate prediction of the SOH of the battery is effectively met under various energy storage scenes, the working state and the health state of the battery are estimated according to the SOH of the battery core, and the safe and stable operation of the energy storage system is ensured. Optionally, the self-adaptive space-time fusion prediction model adopts a PSO-SA hybrid optimization module to combine a particle swarm optimization algorithm and a simulated annealing algorithm to optimize preset structural superparameters in the self-adaptive space-time fusion prediction model, so as to obtain a trained self-adaptive space-time fusion prediction model. According to the technical scheme, the PSO-SA hybrid optimization module is used for effectively improving the super-parameter optimizing precision and the global optimal solving hit rate, enhancing the prediction precision and the space-time characteris