CN-122017609-A - Lithium ion battery life prediction method and system based on state space model

CN122017609ACN 122017609 ACN122017609 ACN 122017609ACN-122017609-A

Abstract

The invention relates to a lithium ion battery life prediction method and system based on a state space model, wherein the method comprises the steps of obtaining characteristic parameters of a battery to be predicted and preprocessing; the method comprises the steps of inputting the preprocessed characteristic parameters into a pre-constructed prediction model to obtain a capacity loss prediction sequence, acquiring charge and discharge cycle data of a lithium ion battery, preprocessing the data to obtain multidimensional input characteristics, performing characteristic conversion to obtain embedded vectors, dividing the embedded vectors into current groups and voltage groups, generating aggregation characteristics and battery internal capacity state variables based on intra-group variable weights and inter-group variable weights, generating time sequence characteristic signals based on the battery internal capacity state variables, performing element-by-element multiplication fusion on the time sequence characteristic signals and original gating signals to obtain time sequence characteristic vectors for prediction, and updating the prediction model based on the capacity loss prediction result. Compared with the prior art, the method has the advantage that the prediction result accords with the physical rule.

Inventors

ZHANG WEINA
ZHANG HAO
LIN FANGZHENG

Assignees

上海电力大学

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. The lithium ion battery life prediction method based on the state space model is characterized by comprising the following steps of: The method comprises the steps of obtaining characteristic parameters including cycle times, capacity and temperature of a battery to be predicted, preprocessing the characteristic parameters, inputting the preprocessed characteristic parameters into a pre-constructed prediction model, and obtaining a capacity loss prediction sequence; The training process of the prediction model comprises the steps of collecting charge-discharge cycle data of a lithium ion battery and preprocessing the charge-discharge cycle data to obtain multidimensional input characteristics, carrying out characteristic conversion on the multidimensional input characteristics to obtain embedded vectors, dividing the embedded vectors into a current group and a voltage group according to physical properties of the input characteristics, respectively carrying out nonlinear conversion on characteristic variables in the voltage group and the current group to generate variable weights in the group, extracting the characteristics of the voltage group and the characteristics of the current group through global average pooling, and further generating voltage group weights and current group weights; Based on the group internal variable weight, the voltage group weight and the current group weight, generating an aggregation characteristic, capturing a dependency relationship in a battery degradation process based on the aggregation characteristic, and obtaining a battery internal capacity state variable; Generating an original gating signal based on the aggregation characteristics, carrying out element-by-element multiplication fusion on the time sequence characteristic signal and the original gating signal to obtain a time sequence characteristic vector, carrying out capacity loss prediction based on the time sequence characteristic vector to obtain a capacity loss prediction result, calculating a loss function comprising a data fitting residual error and a physical residual error based on the capacity loss prediction result, and updating the prediction model until training is converged to obtain a trained prediction model.
2. The method for predicting the life of a lithium ion battery based on a state space model according to claim 1, wherein the physical residuals include degradation rate residuals and capacity loss residuals, and the loss function is expressed as: Wherein, the As a function of the loss, 、 And In order for the coefficient of balance to be present, For the mean square error of the capacity loss prediction, As the mean square error of the degradation rate residual, As the mean square error of the capacity loss residual, For the number of samples to be taken, In order to predict the capacity loss, For real capacity loss, A is a physical model parameter, T is temperature, n is the number of cycles, In order for the degradation rate residual to be a good, As a residual of the capacity loss, In order to achieve a loss of capacity of the battery, Is the factor before the index of the sample, In order for the activation energy to be sufficient, In order to achieve a molar gas constant, In order to provide an attenuation index, As a final capacity loss prediction value, In order to embed the vector(s), Is a neural network parameter.
3. The method for predicting the service life of the lithium ion battery based on the state space model of claim 1, wherein the generating process of the aggregation characteristic comprises the steps of multiplying characteristic variables in a voltage group and a current group by the weights of the variables in the group to respectively obtain the characteristic vectors in the voltage group and the characteristic vectors in the current group; based on the aggregation characteristics, capturing the dependency relationship in the battery degradation process, and performing iterative calculation to obtain the battery internal capacity state variable, wherein the corresponding calculation expression is as follows: Wherein, the To update the battery internal capacity state variables, As a state variable of the internal capacity of the battery, To determine the weight parameters of the historical track retention scale, To control the instantaneous impact weight parameters of the current input, Is an aggregation feature.
4. The method for predicting the life of a lithium ion battery based on a state space model according to claim 3, wherein the state variable of the internal capacity of the battery and the aggregation feature are subjected to linear weighted fusion to obtain a time sequence feature signal, and the corresponding weighted fusion expression is: Wherein, the As a signal of a time sequence characteristic, To influence the projection matrix of states to outputs, Is an offset term of the output.
5. The method for predicting the life of a lithium ion battery based on a state space model according to claim 1, wherein the voltage group characteristics and the current group characteristics are input into a gate residual network, the dynamic contribution of the voltage group and the current group to capacity fading is learned, the voltage group weight and the current group weight are obtained, and the corresponding acquisition expression is: Wherein, the As the feature weights between the groups, For the weights of the voltage group, For the current set weights to be weighted, As a feature of the voltage set, Is characteristic of a current set.
6. The method for predicting the life of a lithium ion battery based on a state space model according to claim 1, wherein feature conversion is performed on the multidimensional input features to obtain an embedded vector, and the corresponding conversion expression is: Wherein, the In order to embed the vector(s), Is a multidimensional input feature.
7. The method for predicting the life of a lithium ion battery based on a state space model according to claim 1, wherein the characteristic variables in the voltage group and the current group are respectively subjected to nonlinear conversion to generate intra-group variable weights, and the corresponding conversion formulas are as follows: Wherein, the For the weights of the variables in the group, For the flattened vectors of all features in the group, For the embedded vector of the j-th set of i-th features, An embedded vector representing the j-th set of features.
8. The method for predicting the life of the lithium ion battery based on the state space model according to claim 1 is characterized by comprising the steps of calculating a termination capacity according to a preset rated capacity after the capacity loss prediction sequence is obtained, further calculating a capacity loss threshold value based on the termination capacity, searching a predicted cycle number corresponding to a prediction sequence which meets the capacity loss threshold value for the first time in the capacity loss prediction sequence, and subtracting the cycle number from the predicted cycle number to obtain the residual service life of the battery.
9. The method for predicting the life of a lithium ion battery based on a state space model according to claim 1, wherein the multidimensional input features specifically comprise a charging time, an accumulated charge amount and charging current and voltage parameters; The charging current and voltage parameters comprise mean value, standard deviation, kurtosis, skewness, curve slope and curve entropy of the charging current and voltage.
10. A system of state space model based lithium ion battery life prediction methods, characterized in that it comprises a memory and a processor, said memory storing a computer program, the processor invoking said computer program to perform the steps of the method according to any of claims 1 to 9.

Description

Lithium ion battery life prediction method and system based on state space model Technical Field The invention relates to the technical field of lithium batteries, in particular to a lithium ion battery life prediction method and system based on a state space model. Background In the current environment-friendly target background, lithium Ion Batteries (LIBs) have become a core component of electric vehicles and energy storage systems by virtue of high energy density and long cycle life. However, during long-term cycling, the battery capacity gradually decays due to internally complex electrochemical side reactions (e.g., SEI film growth, active material loss, etc.). The accurate prediction of the Remaining Useful Life (RUL) of a battery is of vital importance for ensuring safe operation of the system, optimizing maintenance strategies and preventing catastrophic failure. Because the battery degradation process has strong nonlinearity and long time sequence correlation, and is deeply influenced by external conditions such as temperature, running time and the like, the realization of high-precision RUL prediction with physical interpretability is still a great challenge. At present, RUL prediction methods are mainly classified into a mechanism model-based method and a data-driven-based method. Although the mechanism model has strong interpretability, the parameter identification is difficult, the mechanism model is difficult to adapt to complex working conditions, and the generalization capability is limited. The data-driven based method relies on historical operational data training models without deep knowledge of internal mechanisms, where deep learning methods such as LSTM, transformer and the like are prominent in time series data modeling. Chinese patent CN119829956a discloses a prediction method of lithium battery life based on graph neural network, which is related to graph annotation force mechanism capturing characteristics through modal decomposition, chinese patent CN119247194A proposes a prediction method based on Mamba model, and the calculation efficiency is improved by utilizing sequence modeling advantage. However, the existing data driving method still has the defects that most models neglect physical heterogeneity among features, dynamic contribution of different physical attribute features such as voltage, current and the like is difficult to distinguish effectively, part of models lack physical constraint, a prediction result may violate a battery degradation basic rule, physical interpretability is poor, and a traditional time sequence modeling architecture is difficult to consider between long sequence dependency capture and calculation efficiency. The accurate and reliable LIBs RUL prediction technology can provide decision support for battery full life cycle management, and helps the new energy industry to reduce operation cost and improve safety level. The method can promote the engineering application of the battery health management technology by solving the pain points of the existing method in the aspects of physical interpretability, long time sequence modeling precision, characteristic utilization rate and the like, and has important practical significance and industrial value for promoting the high-quality development of the new energy industry. Disclosure of Invention The invention aims to overcome the defects of the prior art and provide a lithium ion battery life prediction method and system based on a state space model. The aim of the invention can be achieved by the following technical scheme: a lithium ion battery life prediction method based on a state space model realizes high-precision and physical credible RUL prediction by fusing a grouping characteristic attention mechanism, mamba time sequence modeling and physical information constraint, and comprises the following steps: The method comprises the steps of obtaining characteristic parameters including cycle times, capacity and temperature of a battery to be predicted, preprocessing the characteristic parameters, inputting the preprocessed characteristic parameters into a pre-constructed prediction model, and obtaining a capacity loss prediction sequence; The training process of the prediction model comprises the steps of collecting charge-discharge cycle data of a lithium ion battery and preprocessing the charge-discharge cycle data to obtain multidimensional input characteristics, carrying out characteristic conversion on the multidimensional input characteristics to obtain embedded vectors, dividing the embedded vectors into a current group and a voltage group according to physical properties of the input characteristics, respectively carrying out nonlinear conversion on characteristic variables in the voltage group and the current group to generate variable weights in the group, extracting the characteristics of the voltage group and the characteristics of the current group through global average pooling, and