Comparative Study of Recurrent Neural Networks for Electric Vehicle Battery State of Charge Estimation

To slow down global climate change, the automotive industry is shifting their attention gradually from internal combustion engine vehicles to electric vehicles. Lithium ion batteries (LiBs) are used to store energy for the propulsion system of electric vehicles. An accurate state-of-charge (SOC) estimator can provide safe and reliable usage of electric vehicles. From the findings in recent literature, it was found that using recurrent artificial neural networks for SOC estimation produce high estimation accuracy. This work investigates two variants of the recurrent neural network studied extensively in the literature, namely, gated recurrent unit (GRU-RNN) and long short-term memory (LSTM-RNN). LiB discharge data when exposed to FUDS, US06 and BJDST driving profiles at 0°C, 25°C and 45°C from CALCE dataset were used for training and evaluating the performance of the ANNs. LSTM-RNN was found to achieve 0.0117% maximum mean squared error (MSE) and GRU-RNN with 0.00975% maximum MSE. However, when tested with real time battery discharge data, it was found that LSTM-RNN is robust to noise than GRU-RNN, thus ensuring more stability in the SOC estimations.