Based On Data-Driven Lithium-Ion Battery SOH And SOC Estimation Research

Battery state of health(SOH)estimation and state of charge(SOC)estimation are important functions of battery management system.In this paper,a data-driven method is used to achieve online high-precision prediction of battery SOH and SOC.Firstly,By analyzing the charging data of the whole battery life cycle,five health indexes related to battery SOH,including constant current charging time,maximum temperature and time to reach maximum temperature,were extracted.The extracted health indicators are detected and repaired to ensure the accuracy of health indicators.The correlation between health indicators and SOH was analyzed to remove indicators with weak correlation.Principal component analysis(PCA)was used to reduce the dimension of the remaining health indicators,eliminate redundant variables,and select two principal components as the input for subsequent SOH estimation.Secondly,the support vector regression(SVR)algorithm is proposed as the core algorithm of SOH estimation to adapt to the problem of less data of two principal components related to SOH.An improved grey wolf optimization algorithm(IGWO)is proposed to optimize the core parameters of SVR to improve the estimation ability of the algorithm.Finally,the IGWO-SVR model is built to determine its core parameters through the training set,and the pre-processed and unpreprocessed data training models are compared,experiments show that the pre-processed data training model has higher prediction accuracy.The trained model is compared with SVR and GWO-SVR in the validation set,which proves that the improved model has better prediction performance.Comparing the prediction of the model in different input data volumes,it is proved that the model can also guarantee better estimation accuracy when the data volume is small.Thirdly,a hybrid model is proposed to estimate SOC by using Mogrifier long-term and short-term memory(LSTM)neural network and convolutional neural network(CNN).Mogrifier long-term and short-term memory(LSTM)neural network is used to extract the time series characteristics of input variables,and convolutional neural network(CNN)is used to extract the relationship between each group of data to fully obtain the characteristics of input data.Finally,the CNN-Mogrifier LSTM hybrid model is built to determine its hyperparameters through the test set,and the trained model is compared with LSTM and Mogrifier LSTM to prove that the hybrid model has better prediction performance.Compared with the commonly used GA-BP network and CNN-GRU,it is proved that the model has higher prediction accuracy and better robustness.Finally,the trained IGWO-SVR program and CNN-Mogrifier LSTM program are embedded in Lab VIEW and used to develop a lithium-ion battery state detection system,which has three functional units : battery data monitoring,battery state prediction and battery fault warning.The battery charge and discharge experimental data were tested to verify its real-time display of battery voltage,current and SOC and early warning display of abnormal state.