State of Charge Estimation for Li-Ion Batteries Based on Recurrent NARX Neural Network with Temperature Effect

The State of Charge (SoC) is a parameter of fundamental importance for the correct operation of the Battery Management Systems (BMS). This data is used in battery charge/discharge control techniques, as well in battery packs load balancing. The forms to determine this parameter usually refer to combinations of two or more estimation methods besides the inclusion of mathematical heuristic or deterministic tools. In this paper, a structure for determining the SoC of lithium-ion batteries based on recurrent nonlinear autoregressive neural networks with external input is presented. The method was developed with the aim of predicting the effects of temperature variation and developing a SoC estimation with low implementation cost, parameters not usually used in the literature. To develop the network and perform validation tests the MATLAB tool was used. In order to verify the efficiency and performance of the proposed neural network, comparative tests with other topologies found in the literature were carried out in addition to noise influence analysis. After the validation tests, a Maximum Mean Error of 1.3920% and a Maximum Average Maximum Error of 5.7759% was obtained.