Lithium Ion Battery State Estimation Using Deep Learning Techniques Utilizing Time Series Data

In the present time, lithium ion batteries are used in varied spheres such as satellites, electric vehicles (EV), solar energy storage and many others. Battery management system (BMS) is necessary to continuously investigate the battery parameters and communicate the same with controlling and communication systems. The BMS is also meant for state estimation of a different kind that is associated with this battery. One such condition is the State of Charge (SoC) which ensures that the battery maintains its efficiency and increases its lifespan. The fact that chemical reactions within these devices are non-linear makes it quite challenging to determine any accurate SoC values which would enable efficient battery management. This paper examines the performance of deep learning RNN models in the context of SOTA(Recursive neural network) and particularly simple RNN, GRU RNN and LSTM when there are additional nonlinearities in output layer, that estimate SoC of lithium ion batteries. As can be deduced from results obtained on train with RMSE 0.029 and test with RMSE 0.033, the LSTM employing leaky RELU as activation function outperforms all other models available.