State of health estimation based on the TimesNet model for real-world electric vehicle batteries

Accurately estimating the state of health (SOH) of power battery is pivotal to enhancing the safety of electric vehicles and realizing the circular utilization of batteries. Given the substandard quality of real-world data and the disparate effects of different regions and operating conditions on vehicle SOH, a novel SOH estimation framework based on data processing methods and TimesNet was proposed herein. Four sets of data from electric vehicles with diverse geographical locations and mileage were preprocessed, calculating the capacity under consideration of the non-linear effect of vehicle charge and discharge patterns and state of charge. Environmental data and real-world data were utilized to extract features, employing density-based spatial clustering of applications with noise clustering and Savitzky–Golay filtering methods to address anomalies and fluctuations in capacity, battery inconsistency features, and charging impedance feature data. Through Spearman correlation analysis, seven inputs were identified, including accumulated mileage, idle days, cell temperature inconsistency, cell voltage inconsistency, charging impedance, and two cumulative weather temperature distribution intervals. The results indicated that the model based on TimesNet achieved the best accuracy compared to other neural network models, demonstrating excellent robustness, with its average absolute percentage error and root mean square error confined to 0.39% and 0.20%.