Abstract: To ensure the accurate and efficient operation of the management system for new energy vehicles, precise evaluation of the state of health(SOH) of power batteries is crucial. Taking into account the influence of ambient temperature and other factors during actual vehicle operation, a data-driven coupled model that considers regional and seasonal temperature conditions is proposed to improve the accuracy of SOH estimation for power battery systems. Vehicles under analysis are divided into two groups based on the average low temperature in winter in their regular operating regions, and their charging behavior is statistically analyzed. Multiple aging characteristic values are selected based on current, voltage, temperature, and incremental capacity curves, and these values are then filtered using the correlation coefficient method. Subsequently, two Bayesian-optimized convolutional neural network(BCNN) models are trained separately using summer and winter data, and their accuracy is validated with test data. Finally, the proposed method is compared with an estimation method based on the long short-term memory neural network(LSTM), and a model coupling optimization scheme is proposed to enhance the adaptability and broad applicability of the estimation method. The results show that the coupled model can achieve an SOH estimation error of no more than 6% for vehicles in low-temperature regions during winter, and an error of no more than 2% for vehicles in warm regions. This indicates that the proposed coupled model can effectively improve the accuracy of SOH estimation under different regional and seasonal temperature conditions.

Key words: lithium-ion battery, state of health estimation, extraction of health factors, real-world driving data, convolutional neural network