State-of-Charge Estimation of Lithium Batteries based on Bidirectional Gated Recurrent Unit and Transfer Learning

Lithium-ion batteries are a vital energy source in contemporary industry, and they have been widely adopted in the automotive power system. This paper proposes a method to estimate the state of charge (SOC) by incorporating a Dropout layer into the bi-directional gated recurrent unit neural network (131GRU), which improves the neural network's generalization capability. This paper examines the effect of temperature factors on the estimated SOC. The network model for estimating SOC at 25 degrees C is rapidly mapped to other temperature conditions using transfer learning shared parameters. Under DST and USO6 operating conditions, both the root means square error and the mean absolute error of the estimated SOC at three ambient temperatures were less than 0.015. And network models obtained through transfer learning can achieve rapid convergence, resulting in a more precise and rapid estimation of SOC at different temperatures.