ata Driven SOH and RUL Early Prediction for Lithium-ion batteries

By reducing the number of required cycles in battery degradation testing, time and costs can be significantly cut. With a data driven approach, an early prediction of state of health and battery lifetime can be achieved. Further, this approach enables features to be extracted and engineered based on measurements from reference performance tests or diagnostic cycles or specific charging or discharging processes during cycling, to then be used as input features in a machine learning model. Our study focuses on early prediction of the state of health (SOH) integrated with the remaining useful life (RUL) of Li-ion batteries using geometrical and statistical features. We employ elastic net and support vector regression (SVR) models within a machine learning pipeline to extract the optimal predictive model. Data collection was conducted at the Stanford Energy Control Laboratory, Stanford University, using INR۲۱۷۰۰-M۵۰T battery cells featuring a graphite/silicon anode and Nickel-Manganese-Cobalt cathode. The tests spanned a ۲۳-month period and utilized the Urban Dynamometer Driving Schedule (UDDS) discharge driving profile, along with a Constant Current (CC)-Constant Voltage (CV) charging protocol at varying rates from C/۴ to ۳C. Our results demonstrate that the proposed algorithm offers several advantages: lower computational cost, higher performance, greater parameter identification stability, and simpler parameter settings. The least MAEP for RUL was ۲.۳% and ۱.۰% for SOH.

کلیدواژه ها:

Li-ion batteries ، SOH prediction ، RUL prediction ، Data-driven early prediction