State-of-Charge and State-of-Energy Estimation for Lithium-Ion Batteries Based on Meta-Learning and Square-Root Unscented Kalman Filter

Simultaneous estimation of state of charge (SOC) and state of energy (SOE) of lithium-ion batteries (LIBs) remains a crucial and challenging issue in the area of new energy electric vehicles, especially in the presence of scarce data. In this article, a method combining a meta-learning scheme and a square-root unscented Kalman filter (meta-learning-SRUKF) is proposed to obtain accurate simultaneous estimations of SOC and SOE. First, the SOC and SOE of the target LIB are estimated through meta-learning, using data from distinct batteries for pretraining and only a small volume of data from the target battery for fine-tuning. Second, an SRUKF is adopted to further smooth the initially estimated SOC and SOE values and reduce the estimation errors. Experiments with various LIBs under distinct drive cycles show that the root-mean-square error (RMSE) and mean absolute error (MAE) for SOC estimation are below 1.6% and 1.5%, respectively. Meanwhile, the RMSE and MAE for SOE are below 1.7% and 1.3%, respectively. Compared with other alternatives, namely transfer learning (TL), gated recurrent unit (GRU), and long short-term memory (LSTM)-based algorithms, the proposed meta-learning-SRUKF approach yields joint estimation of SOC and SOE with higher accuracy and lower complexity.