A novel method for state of health estimation of lithium-ion batteries based on improved LSTM and health indicators extraction

State of health (SOH) is a crucial challenge to guarantee the reliability and safety of the electric vehicles (EVs), due to the complex aging mechanism. A novel SOH estimation method based on improved long short-term memory (LSTM) and health indicators (HIs) extraction from charging-discharging process is proposed in this paper. In order to overcome the limitation of the measurement of battery capacity in real application, some external characteristic parameters related to voltage, current and temperature are selected from charging-discharging process as HIs to describe the aging mechanism of the batteries. After that, Pearson correlation coefficient is employed to select the HIs, which have high correlations with battery capacity. And neighborhood component analysis (NCA) is used to eliminate redundant information of HIs with high correlation in order to reduce computational burden. Aiming at the problem of hyperparameter selection in LSTM models, differential evolution grey wolf optimizer (DEGWO) is proposed in this paper for hyperparameters optimization. Compared with traditional grey wolf optimizer, which is easy to fall into local optimality, DEGWO updates the population through mutation, crossover and screening operations to obtain the global optimal solution and improve the global search ability. The proposed method is verified based on the dataset of the battery from NASA and MIT. The simulations indicate that the proposed method has higher accuracy for different kinds of batteries. The estimation errors for both datasets are within 1%. Compared with other methods, the estimation evaluation indicators such as RMSE, MAE and MAPE of the proposed method are within 1%, which is much less than the estimation results obtained by other methods. And determination coefficient R2 is above 0.95, which means the proposed method has batter fitting performance. It is also indicated that the method proposed in this paper has higher accuracy, better robustness and generalization.(c) 2022 Elsevier Ltd. All rights reserved.

Automated summary

This paper proposes a novel method for state of health (SOH) estimation of electric vehicles (EVs) based on improved long short-term memory (LSTM) and health indicators (HIs) extraction from charging-discharging process. By selecting external characteristic parameters related to voltage, current, and temperature as HIs and using correlation analysis and information elimination methods, the method accurately estimates the battery capacity. Additionally, the proposed method utilizes differential evolution grey wolf optimizer (DEGWO) for hyperparameters optimization, improving the performance of the model. Experimental results using battery datasets from NASA and MIT demonstrate the high accuracy and fitting performance of the proposed method for different types of batteries.