AN EFFECTIVE DEEP LEARNING ALGORITHM BASED ESTIMATION OF STATE OF CHARGE OF LITHIUM ION BATTERY IN HYBRID ELECTRIC VEHICLES

Abstract

The battery pack is the most significant and expensive component in Electric Vehicle, it needs to be closely monitored and controlled. As a result, substantial research is being performed in EV battery state monitoring and control. A neural based method for estimation of SOC is presented in this paper. Two different network structures, i.e., Feed Forward Network and CNN, were considered for investigation. A complete real time data set was trained for the estimation of SOC under various temperature conditions. From the simulation results it can be concluded that instead of BP algorithm, the use of CNN with ‘ADAM’ algorithm provides better training performance, with less number of epochs. The results show that the projected ANN algorithm can provide a good estimate of SOC even in the presence of sensor noise.  In order to improve the life of EV battery, an Adaptive fuzzy controller is proposed for D.C charger module. In this technique  the additional Ultra capacitor reduces the battery degradation.  The average estimation errors of CNN based model under noise disturbance was aways less than 0.1.Hence, proving that the model has strong robustness. Further studies may add the  influence of ageing to optimise the cell SOC estimation, under a variety of operating conditions.