Novel equivalent circuit battery model with adaptive parameters for hybrid state of charge estimation

Due to the increasing demand in electric vehicles (EVs), developments in lithium-ion battery technology are progressing rapidly. As the performance, reliability, and safety of these vehicles depends largely on the battery, it is essential to improve existing battery management systems (BMS) for accurate state estimation. In this contribution, a novel battery cell state estimation technique is developed based on a hybrid and adaptive approach. A long-short term memory (LSTM) model is used for state of charge (SOC) estimation based on experimental data along with an equivalent circuit model (ECM) for simulating the cell dynamics. One unique contribution of this approach lies in hybridizing an ECM with physics-based parameter estimation for real-time applicability and an LSTM-based approach for accurate SOC estimation. Other unique feature is adapting the ECM parameters with aging, dynamic loading, and temperature effect. A further aspect is to integrate a modified Arrhenius equation for ECM parameter estimation with LSTM in a feedback loop along with error compensation to account for accumulation of error. Simulation and experimental results indicate the accuracy of LSTM-based prediction over other models with an error of 4.9 % with training and 1 % with testing dataset of the experimental cell. Results also indicate the impact of cycling on battery aging and SOC estimation. Validation with data from standard database indicates the generalizability of model to other cell types. Finally, the estimated SOC is fed back to adapt the ECM parameters showing an accuracy of 2.1 % when validated with the experimental battery. The feedback loop with error compensation leads to a reduction in accumulated error value by 14.8 % in comparison with other approaches. • In this contribution, a novel battery cell state estimation technique is developed based on a hybrid and adaptive approach • The first objective is to utilize an ECM with physics-based parameter estimation for real-time applicability • Second objective is to utilize the LSTM-based approach for accurate SOC estimation • Simulation and experimental results indicate the accuracy of prediction over other models • Validation with data from standard database indicates the generalizability of model to other cell types