Elman neural network and equivalent circuit model based multi-measurement Kalman filters for SOC estimation

Abstract

This paper provides an insight into the estimation of state of charge (SOC) of lithium-ion batteries, and multi-measurement Kalman filters are constructed to achieve better estimation accuracy. The filter based estimation scheme consists of two sub Kalman filters, one based on the Thevenin equivalent circuit model and the other based on the Elman neural network model. They share the same state equation but have different measurements, namely from the terminal voltage and the Elman neural network, respectively. The optimal weighted error covariance matrix in Kalman filter is computed to obtain the final estimated SOC with the results from two sub filters. The training set of the experiment is from five mixed charging and discharging cycles and the test set is the LA92 driving cycle. The experimental results show that multi-measurement Kalman filtering has higher accuracy in SOC estimation, both the mean absolute error and the root mean square error are less than 0.3\%.