State of Charge Estimation for Lithium-ion Battery Based on Random Forests Technique with Gravitational Search Algorithm

Abstract

An accurate state of charge (SOC) estimation for lithium-ion battery has been an intensively researched subject in electric vehicle (EV) application towards the advancement of the sustainable transportation system. However, SOC estimation with high accuracy is challenging because of the complex internal characteristics of the lithium-ion battery which is changed by different environmental situations. This paper develops an accurate method for the state of charge (SOC) estimation of a lithium-ion battery using random forests (RFs) algorithm. However, the accuracy of RFs highly depends on the appropriate selection of trees and leaves per tree in a forest. Thus, this research develops an enhanced model with RFs based gravitational search algorithm (GSA). The aim of GSA is to find the best value of trees and leaves per tree. The robustness and accuracy of the proposed model are tested under different temperatures. The model training and validation are executed using federal urban driving schedule (FUDS). The effectiveness of the proposed method is compared with the conventional RFs and radial basis function neural network (RBFNN) and optimal RBFNN-GSA models using different statistical error terms and computational cost. The proposed RFs based GSA model offers higher robustness and accuracy in reducing RMSE by 55.4%, 67.4%, and MAE by 39.1% and 78.1% than conventional RFs and RBFNN based GSA model, respectively at 25°C.