CN-117406121-B - SOH estimation method for lithium ion battery of energy storage power station

CN117406121BCN 117406121 BCN117406121 BCN 117406121BCN-117406121-B

Abstract

The invention relates to an SOH estimation method of a lithium ion battery of an energy storage power station, which comprises the steps of collecting relaxation voltage data and SOH values of the lithium ion battery after full charge, extracting a plurality of characteristics and calculating first-order differential voltage data from the collected data, dividing a dataset into a training set and a testing set, constructing LightGBM models, respectively adopting the training set and the testing set to train and test the models, constructing a one-dimensional convolutional neural network CNN, respectively adopting the training set and the testing set to train and test the models, combining SOH values estimated by the extracted characteristics, a LightGBM model and a CNN model into a characteristic dataset, constructing a linear regression LR model, training and testing the characteristic dataset by utilizing the characteristic dataset, and inputting SOH values estimated by the extracted characteristics, a LightGBM model and the CNN model into the LR model to obtain a final SOH estimation result. The method has good generalization capability and high estimation precision.

Inventors

HUANG XINGHUA
HE FENG
LIANG ZIKANG
CHEN SIZHE
ZHENG YU
FAN YUANLIANG
WU HAN
CHEN KUOSONG
ZHU JUNWEI
LI ZEWEN
CHEN WEIMING
LIN JIANLI
LI LINGFEI

Assignees

国网福建省电力有限公司
国网福建省电力有限公司电力科学研究院
国网福建省电力有限公司莆田供电公司

Dates

Publication Date: 20260508
Application Date: 20230928

Claims (6)

1. The SOH estimation method of the lithium ion battery of the energy storage power station is characterized by comprising the following steps of: S1, collecting relaxation voltage sequence data V of a lithium ion battery in a period of 5 minutes after the lithium ion battery is fully charged in each cycle and SOH label values of the battery in each cycle, and forming an original data set D 1 ; S2, calculating a maximum value V max , a minimum value V min , shannon entropy V entropy , an average value V mean and a standard deviation value V std of relaxation voltage sequence data in each cycle as characteristics according to the original data set D 1 obtained in the step S1, forming a characteristic data set D f,1 with SOH label values, calculating first-order differential voltage sequence data V of the relaxation voltage sequence data, and forming a characteristic data set D f,2 by the relaxation voltage sequence data V, the first-order differential voltage sequence data V and corresponding SOH label values; S3, carrying out data set division on the characteristic data sets D f,1 and D f,2 in the same division mode to obtain training sets D train,1 and D train,2 and test sets D test,1 and D test,2 respectively; S4, constructing LightGBM a model, respectively adopting a training set D train,1 and a testing set D test,1 to train and test the LightGBM model, adopting an average absolute error MAE to measure the estimation precision of the LightGBM model, and when the MAE of the model on the testing set is less than or equal to 2%, completing training by the model, otherwise, repeating the step S4 until the condition is met; S5, constructing a one-dimensional convolutional neural network CNN, respectively adopting a training set D train,2 and a testing set D test,2 to train and test the CNN model, adopting an average absolute error MAE to measure the estimation precision of the CNN model, and when the MAE of the model on the testing set is less than or equal to 2%, completing training by the model, otherwise, repeating the step S5 until the condition is met; S6, combining the 5 relaxation voltage characteristics extracted in the step S2, namely a maximum value V max , a minimum value V min , shannon entropy V entropy , an average value V mean and a standard deviation value V std , an SOH estimation result output by a LightGBM model on a characteristic data set D f,1 and an SOH estimation result output by a CNN model on a characteristic data set D f,2 , forming a characteristic data set D f,3 by the SOH estimation result and SOH label values, and dividing the characteristic data set D f,3 into data sets according to a dividing mode of the step S3 to obtain a training set D train,3 and a test set D test,3 ; S7, constructing an LR model, respectively adopting a training set D train,3 and a testing set D test,3 to train and test the LR model, similarly, adopting an average absolute error MAE to measure the estimation precision of the linear regression LR model, and when the MAE of the model on the testing set is less than or equal to 1.5%, completing training by the model, otherwise, repeating the step S7 until the condition is met; s8, importing the LightGBM, CNN and LR models obtained through training in the steps S4, S5 and S7 into a battery management system, extracting characteristics and calculating first-order differential voltage sequence data according to the acquired relaxation voltage sequence data in a mode of the step S2 after a lithium ion battery in an energy storage power station is charged and stands for 5 minutes, inputting 5 relaxation voltage characteristics into the LightGBM models to obtain an SOH estimation result, inputting original relaxation voltage sequence data and first-order differential voltage sequence data into the CNN models to obtain an SOH estimation result, and inputting the SOH estimation result output by the two models and 5 relaxation voltage characteristics into the LR models together to obtain a final SOH estimation result; the step S2 specifically comprises the following steps: S2-1, calculating a maximum value V max , a minimum value V min , shannon entropy V entropy , an average value V mean and a standard deviation value V std of relaxation voltage sequence data in each cycle as features according to the original data set D 1 obtained in the step S1, forming a feature data set D f,1 with SOH label values, and carrying out normalization processing, wherein the specific form is as follows: Wherein, the Representing the maximum value of the relaxation voltage in the ith cycle, Representing the minimum value of the relaxation voltage in the i-th cycle, The shannon entropy value representing the relaxation voltage in the ith cycle is calculated as , Representing the probability of the value of the nth element in the data sequence, N representing the total N values in the data sequence, Representing the average value of the relaxation voltage in the i-th cycle, Representing the standard deviation of the relaxation voltage in the ith cycle; s2-2, calculating first-order differential voltage sequence data fatin V of relaxation voltage sequence data, wherein the specific form of the sequence data of the ith cycle is as follows: Wherein, the The relaxation voltage sequence data V, the first-order differential voltage sequence data fatin V and the corresponding SOH label value form a characteristic data set D f,2 , and normalization processing is carried out, wherein the specific form is as follows: 。
2. the SOH estimation method of a lithium ion battery of an energy storage power station according to claim 1, wherein the step S1 specifically comprises the steps of: S1-1, collecting relaxation voltage sequence data V of the lithium ion battery during standing for 5 minutes after full charge in each cycle, wherein the specific form of the sequence data of the ith cycle is as follows: Wherein, the An nth voltage value representing an ith cycle; s1-2, the relaxation voltage sequence data V and the SOH label value of the battery in each cycle form an original data set D 1 , and the specific form is as follows: 。
3. the SOH estimation method of a lithium ion battery of an energy storage power station according to claim 1, wherein the step S4 specifically comprises the steps of: S4-1, constructing LightGBM models, and setting the learning rate, the number of estimators and the maximum depth of the tree of the models; S4-2, training and testing the LightGBM model by adopting the training set D train,1 and the testing set D test,1 obtained in the step S3, and measuring the estimation accuracy of the LightGBM model by adopting an average absolute error MAE, wherein the calculation formula of the MAE is as follows: Wherein, the Representing the SOH estimation result of the battery at the first cycle of the model output, An actual SOH value representing the battery at the first cycle; When LightGBM the MAE of the model on the test set is less than or equal to 2%, the model completes training, otherwise, the step S4 is repeated until the condition is met.
4. The SOH estimation method of a lithium ion battery of an energy storage power station according to claim 1, wherein the step S5 specifically comprises the steps of: S5-1, constructing a one-dimensional convolutional neural network CNN, and setting the number of layers, the number of convolutional kernels, the size of the convolutional kernels, the moving step length of the convolutional kernels and the learning rate of the model; And S5-2, training and testing the CNN model by adopting the training set D train,2 and the testing set D test,2 obtained in the step S3, measuring the estimation precision of the CNN model by adopting the average absolute error MAE, and when the MAE of the model on the testing set is less than or equal to 2%, completing training by the model, otherwise, repeating the step S5 until the condition is met.
5. The SOH estimation method of a lithium ion battery of an energy storage power station according to claim 1, wherein the step S6 specifically comprises the steps of: S6-1, combining the 5 relaxation voltage characteristics extracted in the step S2, namely a maximum value V max , a minimum value V min , shannon entropy V entropy , an average value V mean and a standard deviation value V std , and the SOH estimation result output by the LightGBM model on the characteristic dataset D f,1 and the SOH estimation result output by the CNN model on the characteristic dataset D f,2 , and combining the relaxation voltage characteristics with SOH label values to form a characteristic dataset D f,3 , wherein the specific form is as follows: Wherein, the Representing the SOH estimate output by the LightGBM model in the ith cycle, Representing an SOH estimated value output by the CNN model in the ith cycle; And S6-2, dividing the characteristic data set D f,3 according to the dividing mode of the step S3 to obtain a training set D train,3 and a testing set D test,3 .
6. The SOH estimation method of a lithium ion battery of an energy storage power station according to claim 1, wherein the step S7 specifically comprises the steps of: S7-1, constructing a linear regression LR model; And S7-2, training and testing the LR model by adopting the training set D train,3 and the testing set D test,3 obtained in the step S6, measuring the estimation accuracy of the LR model by adopting the average absolute error MAE, and when the MAE of the LR model on the testing set is less than or equal to 1.5%, completing training by the model, otherwise, repeating the step S7-2 until the condition is met.

Description

SOH estimation method for lithium ion battery of energy storage power station Technical Field The invention relates to the technical field of battery energy storage, in particular to an SOH estimation method of a lithium ion battery of an energy storage power station. Background With the recent trend of exhaustion of traditional energy and the increasing deterioration of ecological environment, research and development of new energy have been receiving a great deal of attention. As one of the new energy types, lithium ion batteries are gradually widely used in the fields of power grid energy storage, electric automobiles, aerospace and the like. Over time, the health life SOH of lithium ion batteries can drop irreversibly. When the SOH of the battery is lower than a certain threshold, the energy storage performance of the battery may decrease and the occurrence rate of failure may increase. Therefore, accurate estimation of SOH of a battery is of great importance for safe use of the battery. The existing SOH estimation method of the lithium ion battery has stricter requirements on the operation condition of the lithium ion battery, for example, the method based on a part of constant-current charging curve completely passes through a set voltage window when the battery is required to be charged, the method based on the constant-voltage charging curve can be influenced by the magnitude of charging current, and the method based on the discharging voltage curve is difficult to be applied when facing the actual operation scene. In addition, the SOH estimation method based on data driving only considers the mode of manually extracting the characteristics or automatically extracting the characteristics by a deep learning model. However, both feature extraction approaches have their own features and advantages, and are not in a contradictory relationship. Therefore, there is a need to develop an SOH estimation method that has low requirements on battery operating conditions and combines manual extraction and automatic extraction of features. Disclosure of Invention The invention aims to provide an SOH estimation method of a lithium ion battery of an energy storage power station, which has the advantages of good generalization capability and high estimation precision. In order to achieve the purpose, the technical scheme adopted by the invention is that the SOH estimation method of the lithium ion battery of the energy storage power station comprises the following steps: S1, collecting relaxation voltage sequence data V of a lithium ion battery in a period of 5 minutes after the lithium ion battery is fully charged in each cycle and SOH label values of the battery in each cycle, and forming an original data set D 1; S2, calculating a maximum value V max, a minimum value V min, shannon entropy V entropy, an average value V mean and a standard deviation value V std of relaxation voltage sequence data in each cycle as characteristics according to the original data set D 1 obtained in the step S1, forming a characteristic data set D f,1 with SOH label values, calculating first-order differential voltage sequence data V of the relaxation voltage sequence data, and forming a characteristic data set D f,2 by the relaxation voltage sequence data V, the first-order differential voltage sequence data V and corresponding SOH label values; S3, carrying out data set division on the characteristic data sets D f,1 and D f,2 in the same division mode to obtain training sets D train,1 and D train,2 and test sets D test,1 and D test,2 respectively; S4, constructing LightGBM a model, respectively adopting a training set D train,1 and a testing set D test,1 to train and test the LightGBM model, adopting an average absolute error MAE to measure the estimation precision of the LightGBM model, and when the MAE of the model on the testing set is less than or equal to 2%, completing training by the model, otherwise, repeating the step S4 until the condition is met; S5, constructing a one-dimensional convolutional neural network CNN, respectively adopting a training set D train,2 and a testing set D test,2 to train and test the CNN model, adopting an average absolute error MAE to measure the estimation precision of the CNN model, and when the MAE of the model on the testing set is less than or equal to 2%, completing training by the model, otherwise, repeating the step S5 until the condition is met; S6, combining the 5 relaxation voltage characteristics extracted in the step S2, namely a maximum value V max, a minimum value V min, shannon entropy V entropy, an average value V mean and a standard deviation value V std, an SOH estimation result output by a LightGBM model on a characteristic data set D f,1 and an SOH estimation result output by a CNN model on a characteristic data set D f,2, forming a characteristic data set D f,3 by the SOH estimation result and SOH label values, and dividing the characteristic data set D f,3 into data sets ac