SOC estimation for lithium-ion battery using the LSTM-RNN with extended input and constrained output

Abstract

The state of charge (SOC) estimation of lithium-ion battery (LIB) based on recurrent neural network (RNN) has been a popular research due to its suitability for time series data prediction. However, there are significant output fluctuations in solo network, which lead to unstable SOC estimation performance. To solve this problem, this paper proposes a novel long short-term memory recurrent neural network (LSTM-RNN) with extended input (EI) and constrained output (CO) for battery SOC estimation, named EI-LSTM-CO. For the network input, an additional slow time-varying information sliding window average voltage is introduced to enhance the ability of network to map the nonlinear characteristics of the battery and reduce the output SOC fluctuations. In terms of the network output, a state flow strategy based on the Ampere-hour integration (AhI) is designed to constrain the variation between adjacent output SOCs of the network to smooth the network output and further improve the SOC estimation performance. In the experiments, the LiFePO4 battery datasets at various temperatures are used to validate the SOC estimation performance and generalization ability. In particular, the root mean square error (RMSE) and the maximum error (MAXE) of the proposed method on unknown data are less than 1.3% and 3.2% respectively.