CN-121978572-A - Lithium ion battery health state estimation method and system

CN121978572ACN 121978572 ACN121978572 ACN 121978572ACN-121978572-A

Abstract

The invention discloses a lithium ion battery health state estimation method and system, and relates to the technical field of battery management, comprising the following steps of collecting a current-time curve of a target lithium ion battery in a constant voltage charging stage, and removing an unstable section of the current-time curve to obtain a stable section; randomly selecting an initial current point in a stable interval of a current-time curve, intercepting a current segment with set time length from the initial point, resampling the current segment according to a preset sampling interval to obtain a current characteristic sequence, obtaining a first derivative sequence corresponding to the current characteristic sequence, and inputting the current characteristic sequence and the first derivative sequence thereof into a pre-trained SOH estimation model to obtain a corresponding SOH estimation value. The invention realizes the data high-efficiency SOH estimation with good precision and generalization under different charging strategies.

Inventors

GAO FENG
PAN WENJIA
WANG SHUQUAN
TIAN HAO

Assignees

山东大学

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (9)

1. The lithium ion battery health state estimation method is characterized by comprising the following steps of: Collecting a current-time curve of a target lithium ion battery in a constant voltage charging stage, and removing an unstable section of the current-time curve to obtain a stable section; Randomly selecting an initial current point in a stable interval of a current-time curve, intercepting a current segment with set time length from the initial point backwards, and resampling the current segment according to a preset sampling interval to obtain a current characteristic sequence; and inputting the current characteristic sequence and the first derivative sequence thereof into a pre-trained SOH estimation model to obtain a corresponding SOH estimation value.
2. The method for estimating the health state of a lithium ion battery according to claim 1, wherein the step of removing the unstable section of the current-time curve to obtain the stable section comprises the following steps: Resampling the current-time curve to obtain a reconstructed current curve; carrying out smooth filtering on the reconstructed current curve to obtain a smooth current curve; and eliminating part of the sections of the smooth current curve based on the set amplitude threshold value to obtain a stable current section.
3. The method for estimating the health state of a lithium ion battery according to claim 1, wherein the SOH estimation model comprises a transducer network, a one-dimensional convolutional neural network, a connecting layer and a regression layer, wherein the transducer network comprises a linear embedding layer, a position coding layer, a multi-head attention mechanism, a feedforward neural network and a linear layer, the one-dimensional convolutional neural network comprises a plurality of convolutional layers, a pooling layer and a full-connection layer, and an activation function is arranged behind each convolutional layer; The linear embedding layer maps the current characteristic sequence to obtain a vector sequence, the position coding layer performs position coding on the vector sequence, adds the position coding of the vector sequence and the vector sequence to obtain an input sequence, models the dependency relationship between different time positions of the input sequence by a multi-head attention mechanism to obtain an input characteristic matrix, and maps the input characteristic matrix by a feedforward neural network to obtain a sequence characteristic matrix; the convolution layers carry out convolution on the first derivative sequence, and the pooling layer carries out aggregation on the convolution result of the last layer; The connection layer splices the first embedded vector and the second embedded vector to obtain a joint feature representation, and the regression layer maps the joint feature representation to obtain an SOH estimated value.
4. The method of claim 1, wherein the pre-training of the SOH estimation model comprises the steps of: Charging and discharging cycles are carried out on lithium ion batteries of the same type under different charging protocols, wherein the charging and discharging cycles comprise conventional cycles and diagnosis cycles, and the diagnosis cycles all adopt the same charging and discharging strategies under different charging protocols; For charge and discharge cycles under each charge protocol, SOH values corresponding to a plurality of diagnosis cycles are obtained, and SOH values of a plurality of conventional cycles are interpolated through the SOH values to obtain a continuous SOH label sequence of the whole life cycle of the lithium ion battery under the current charge protocol; For the conventional cycle under each charging protocol, collecting a current-time curve in a constant voltage charging stage, and removing an unstable section of the current-time curve to obtain a corresponding stable section; selecting a plurality of initial current levels in a stable interval according to a set current amplitude value, acquiring an initial current value at each initial current level, intercepting a current segment with a set time length backwards, and resampling the current segment according to a preset sampling interval to obtain a current characteristic sequence; And training the SOH estimation model through the training set.
5. A method of estimating the state of health of a lithium-ion battery according to claim 2, wherein the current-time curve is resampled using cubic spline interpolation, the resampling time step being 0.1 seconds.
6. The method of claim 2, wherein the stable current interval is obtained by eliminating a segment with a starting current greater than 1.0A and a segment with a terminal current less than 0.25A.
7. The method of claim 1, wherein the current signature sequence is 20 sampling points, each sampling point corresponding to a1 second sampling interval.
8. The method for estimating a state of health of a lithium ion battery according to claim 1, wherein the set time period is 10s to 60s.
9. A lithium ion battery state of health estimation system, comprising: The acquisition module is used for acquiring a current-time curve of the target lithium ion battery in a constant voltage charging stage, and eliminating an unstable section of the current-time curve to obtain a stable section; The extraction module is used for randomly selecting an initial current point in a stable interval of a current-time curve, intercepting a current segment with set time length from the initial point, and resampling the current segment according to a preset sampling interval to obtain a current characteristic sequence; and the estimation module is used for inputting the current characteristic sequence and the first derivative sequence thereof into a pre-trained SOH estimation model to obtain a corresponding SOH estimation value.

Description

Lithium ion battery health state estimation method and system Technical Field The invention relates to the technical field of battery management, in particular to a lithium ion battery health state estimation method and system. Background The lithium ion battery has the advantages of high energy density, low self-discharge rate, long cycle life and the like, and is widely applied to electric automobiles and energy storage systems. However, as the cycle number increases, aging phenomena such as capacity fade and internal resistance increase of the lithium ion battery inevitably occur, resulting in gradual degradation of performance and potential safety hazard. Therefore, how to accurately and efficiently evaluate the State Of Health (SOH) Of the lithium battery has important significance for guaranteeing the safety and reliability Of the battery system. The existing SOH estimation method mainly comprises a model driving method and a data driving method. The model driving method characterizes the battery behavior by constructing an equivalent circuit model, an electrochemical model or an empirical degradation model, and generally needs to collect continuous data for a long time under controlled conditions and perform complex mechanism modeling and parameter identification, and model parameters need to be dynamically updated along with battery aging, so that the calculation and realization cost is high, and the method is difficult to adapt to actual application scenes with changeable working conditions and obvious individual differences. The data driving method utilizes the voltage and current curve characteristics in the charge and discharge process, combines a machine learning or deep learning model to carry out SOH estimation, and has the advantages of strong flexibility and simple modeling. It has been shown that typical health indexes such as differential voltage (DIFFERENTIAL VOLTAGE, DV) and incremental capacity (INCREMENTAL CAPACITY, IC) can obtain higher estimation accuracy even if matched with a basic machine learning model, but complete charging data over 30 minutes are generally needed, in order to shorten the characteristic acquisition time, a method is provided for only extracting the charging capacity or the charging time in a specific voltage interval as a health index and reducing the sampling time to about 10 minutes, and further work is performed for shortening the characteristic acquisition window to about 3 minutes by extracting the voltage increment in a high voltage interval. Although the above studies have improved the data acquisition efficiency to some extent, continuous stable charging segments of several minutes are still required as a precondition. In practical situations, the lithium ion battery is often in a condition of being charged frequently and randomly in different charge states for a short time, so that the premise is often difficult to meet, unstable feature extraction and limited data utilization rate are caused, and the accuracy of SOH estimation of the lithium ion battery is affected. Disclosure of Invention Based on the defects in the prior art, the invention provides a lithium ion battery health state estimation method and system, which solve the problems that the existing method depends on long-time charging data, so that feature extraction is unstable, the data utilization rate is limited, and the SOH estimation accuracy of a lithium battery is affected. The invention adopts the following technical scheme: in a first aspect, the present invention provides a method for estimating a state of health of a lithium ion battery, comprising the steps of: Collecting a current-time curve of a target lithium ion battery in a constant voltage charging stage, and removing an unstable section of the current-time curve to obtain a stable section; Randomly selecting an initial current point in a stable interval of a current-time curve, intercepting a current segment with set time length from the initial point backwards, and resampling the current segment according to a preset sampling interval to obtain a current characteristic sequence; and inputting the current characteristic sequence and the first derivative sequence thereof into a pre-trained SOH estimation model to obtain a corresponding SOH estimation value. Preferably, the removing the unstable section from the current-time curve to obtain a stable section specifically includes the following steps: Resampling the current-time curve to obtain a reconstructed current curve; carrying out smooth filtering on the reconstructed current curve to obtain a smooth current curve; and eliminating part of the sections of the smooth current curve based on the set amplitude threshold value to obtain a stable current section. Preferably, the SOH estimation model comprises a transducer network, a one-dimensional convolutional neural network, a connecting layer and a regression layer, wherein the transducer network comprises a linear embeddi