CN-122020414-A - Battery health state assessment method and system

CN122020414ACN 122020414 ACN122020414 ACN 122020414ACN-122020414-A

Abstract

The invention provides a battery health state assessment method and system, which comprise the steps of obtaining multiple groups of training data and corresponding health states, generating corresponding enhancement state data by using a GAN model according to each group of training data, constructing a cross feature model, training the cross feature model by using the enhancement state data to obtain an optimized feature model, inputting all enhancement state data into the optimized feature model to obtain multiple enhancement feature data, classifying the enhancement feature data by using a clustering algorithm to obtain multiple state classifications, obtaining original state data of a current battery, inputting the original state data into the optimized assessment model to obtain the original feature data, and classifying the original feature data by using the clustering algorithm to obtain the corresponding state classifications. The method solves the problems that a large number of labeling samples are required to train a model and information loss exists in the prior art when the data features of the labeling samples are extracted.

Inventors

QIU ZHAO
TIAN SHIBAO
TU JIANGFENG
ZENG QIN
CAO JING

Assignees

重庆标能瑞源储能技术研究院有限公司

Dates

Publication Date: 20260512
Application Date: 20251217

Claims (9)

1. A battery state of health assessment method is characterized by comprising the following steps: acquiring a plurality of groups of training data and corresponding health states, and generating corresponding enhancement state data by using a GAN model according to each group of training data; Constructing a cross feature model, training the cross feature model by using the enhancement state data to obtain an optimized feature model, and then inputting all enhancement state data into the optimized feature model to obtain a plurality of enhancement feature data; classifying the enhanced feature data by using a clustering algorithm to obtain a plurality of state classifications, wherein the state classifications comprise a plurality of enhanced feature data and corresponding health states; The method comprises the steps of obtaining original state data of a current battery, inputting the original state data into an optimized evaluation model to obtain original feature data, classifying the original feature data by using a clustering algorithm to obtain corresponding state classifications, and judging the health state of the current battery according to the quantity of enhanced feature data corresponding to each health state in the state classifications.
2. A battery state of health assessment method as defined in claim 1, wherein: The method for generating corresponding enhancement state data by using the GAN model according to each group of training data comprises the following steps: Positioning the position of an abnormal value in the training data by using an isolated forest algorithm, and marking; constructing a sliding window, extracting training data with a fixed sliding step length to obtain a plurality of data subsequences, and then eliminating all data subsequences with abnormal values; And constructing a GAN model, training the GAN model by using all data subsequences, and generating enhancement state data by using the trained GAN model.
3. A battery state of health assessment method as defined in claim 1, wherein: Training the cross feature model by using the enhancement state data, and obtaining the optimized feature model comprises the following steps: s1, dividing enhancement state data into first input data and second input data according to a random proportion; S2, respectively inputting the first input data and the second input data into a cross feature model to obtain a first data feature and a second data feature; S3, after calculating the feature absolute difference between the first data feature and the second data feature, inputting the feature absolute difference into a fully-connected neural network to carry out similarity judgment, and adjusting parameters of the cross feature model according to a judgment result; and S4, repeating the steps S1-S3 until the cross feature model converges to obtain an optimized feature model.
4. A battery state of health assessment method as defined in claim 1, wherein: All the enhancement state data are input into an optimized feature model, and the method for obtaining a plurality of enhancement feature data comprises the following steps: S1, equally dividing the enhancement state data into a plurality of segment data, and respectively carrying out feature extraction on each segment data by adopting a plurality of same enhancement feature extraction networks to obtain a plurality of segment feature data; S2, dividing each piece of sectional feature data equally, and carrying out cross combination on all pieces of sectional feature data according to the equally divided data positions of the sectional feature data to obtain a plurality of cross feature data; s3, carrying out feature addition on all the cross feature data to obtain corresponding dimension reduction data; S4, taking the dimensionality reduction data as enhancement state data, repeating the steps S1-S4 until the maximum iteration times are reached, and taking the dimensionality reduction data obtained in the last iteration as corresponding enhancement characteristic data; And S5, traversing all the enhancement state data, and repeating the steps S1-S4 to obtain enhancement characteristic data corresponding to each enhancement state data.
5. A battery state of health assessment method as defined in claim 1, wherein: In S4, when each iteration is performed, the number of the segmented data obtained by segmenting the enhancement state data is different from the number in all previous iteration processes.
6. The battery state of health assessment method according to claim 5, wherein: the enhanced feature extraction network comprises three layers of one-dimensional convolution networks, wherein the input channel of the first layer of one-dimensional convolution network is 1, the output channel of the first layer of one-dimensional convolution network is n, the input channel of the second layer of one-dimensional convolution network is n, the output channel of the second layer of one-dimensional convolution network is n, the input channel of the third layer of one-dimensional convolution network is n, and the output channel of the third layer of one-dimensional convolution network is 1.
7. A battery state of health assessment method as defined in claim 1, wherein: And counting the quantity of the enhancement feature data corresponding to each health state in the state classification of the original state data, and taking the health state corresponding to the enhancement feature data with the largest quantity as the health state of the original state data.
8. The battery state of health assessment method according to claim 7, wherein: If the quantity of training data corresponding to multiple health states is the same in the state classification corresponding to the original state data, the clustering model is used for reclustering the original characteristic data on the premise of eliminating the state classification.
9. A battery state of health assessment system, characterized in that the system uses a battery state of health assessment method according to any one of claims 1-8, comprising: the data acquisition module is used for acquiring training data, corresponding health states and original state data of the current battery; The model building module is used for building and training a GAN model and a cross feature model; the classification module is used for classifying the enhanced feature data and the original feature data by using a clustering algorithm; the judging module is used for judging the current health state of the battery according to the quantity of the enhanced characteristic data corresponding to each health state in the belonging state classification.

Description

Battery health state assessment method and system Technical Field The present invention relates to the field of battery status monitoring technologies, and in particular, to a method and a system for evaluating a battery health status. Background With the rapid development of the global electric automobile industry in recent years, the holding capacity of electric automobiles is continuously increased, and the power battery is taken as an important component part in the three-electric system of the electric automobiles, and the cost of the power battery is about one third of the cost of the whole automobile. The health of the power battery directly determines the performance and the residual value of the vehicle, and is a key factor affecting the user experience of the electric automobile. The current detection of the state of health of the power battery is mainly finished by professional battery testing equipment, key parameters such as capacity and impedance of the battery are collected in the process of charging and discharging the battery, indexes such as SOC and SOH of the power battery are calculated through a trained calculation model to judge the state of health of the power battery, and the method is accurate in evaluation of the state of health of the battery, but the calculation model needs a large number of labeling samples for supporting. In addition, the calculation model usually uses a convolutional neural network to extract the data characteristics of the parameters, but the convolutional neural network only keeps the summarized information such as the maximum value or the average value in a pooling window during extraction, discards specific characteristics and position information, has the defect of information loss, and can cause inaccurate final calculation results. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides a battery health state evaluation method and a system, which solve the problems that a large number of labeling samples are required to train a model and information loss exists in the data characteristics of the extracted labeling samples in the prior art. According to an embodiment of the present invention, a battery state of health evaluation method includes: acquiring a plurality of groups of training data and corresponding health states, and generating corresponding enhancement state data by using a GAN model according to each group of training data; Constructing a cross feature model, training the cross feature model by using the enhancement state data to obtain an optimized feature model, and then inputting all enhancement state data into the optimized feature model to obtain a plurality of enhancement feature data; classifying the enhanced feature data by using a clustering algorithm to obtain a plurality of state classifications, wherein the state classifications comprise a plurality of enhanced feature data and corresponding health states; The method comprises the steps of obtaining original state data of a current battery, inputting the original state data into an optimized evaluation model to obtain original feature data, classifying the original feature data by using a clustering algorithm to obtain corresponding state classifications, and judging the health state of the current battery according to the quantity of enhanced feature data corresponding to each health state in the state classifications. Preferably, the method for generating corresponding enhancement state data using the GAN model according to each set of training data comprises: Positioning the position of an abnormal value in the training data by using an isolated forest algorithm, and marking; constructing a sliding window, extracting training data with a fixed sliding step length to obtain a plurality of data subsequences, and then eliminating all data subsequences with abnormal values; And constructing a GAN model, training the GAN model by using all data subsequences, and generating enhancement state data by using the trained GAN model. Preferably, the method for training the cross feature model by using the enhancement state data to obtain the optimized feature model comprises the following steps: s1, dividing enhancement state data into first input data and second input data according to a random proportion; S2, respectively inputting the first input data and the second input data into a cross feature model to obtain a first data feature and a second data feature; S3, after calculating the feature absolute difference between the first data feature and the second data feature, inputting the feature absolute difference into a fully-connected neural network to carry out similarity judgment, and adjusting parameters of the cross feature model according to a judgment result; and S4, repeating the steps S1-S3 until the cross feature model converges to obtain an optimized feature model. Preferably, the method for inputting all enhancement state data into