State of health analysis of batteries at different stages based on real-world vehicle data and machine learning

The capacity and performance of batteries decay over time. How to deal with retired batteries is a major challenge at present. This study proposes a health state assessment method for retired batteries. The Forgetting Factor Recursive Least Square is used for parameter identification based on the operation data of new energy vehicles at different mileage periods. Ohmic internal resistance is extracted and used as a characteristic parameter to characterize the state of health. The internal resistances of the vehicles at different driving cycles are compared and their variations are derived. Parameters highly correlated with the battery ohmic internal resistance are selected as input parameters for the long and short-term memory neural network. The accurate state of health prediction model is obtained after parameter tuning. The root-mean-square error of the predicted results is less than 0.01 Ω. This shows that the proposed method can effectively assess the power battery state of health. The study can provide a basis for the stepwise utilization of retired batteries, and thus promote the development of sustainable battery utilization. •State of health analysis is essential to battery operation and retired battery assessment.•Equivalent circuit model is used to characterize the reaction processes inside the battery.•Forgetting factor recursive least square voltage is used to recognize the internal resistance.•A novel state of health prediction strategy based on machine learning is proposed.•Prediction effects are verified using data for different vehicles and different mileage.