Hybrid Adaptive Filtering Approaches for Lithium-Ion Battery State of Charge and Health Estimation

Abstract

Approach to a hybrid algorithm that would combine Recurrent Neural Networks, Kalman Filters, and estimates the State of Charge and Health of lithium-ion batteries. Normally, the established methods fail while dealing with their non-linear characteristics and dynamic operation, thus sometimes giving erroneous predictions. The proposed Hybrid Kalman Filter (HKF) combines strengths of RNNs and Kalman filters. RNNs are used as they can well model complex temporal dependencies and non-linear relationships in the battery data that improve the Kalman filter prediction capabilities. This algorithm works under two main stages: training the RNN on historical data with the goal to learn the battery dynamics and exploit these insights in real-time estimation of SOC and SOH. The experimental validation also proved that the HKF performs superiorly than other conventional methods such as UKF, especially concerning the lower values of RMSE achieved under changing conditions of C-rate (slow and fast charge/discharge rates). This is what ensures the efficient management of a battery in better performance, safety, and durability. It does have great promise for use in electric vehicles, renewable energy systems, and portable electronics where accurate battery monitoring is important to ensure reliable and efficient operation.