Deep Learning based EV Battery Management System

The invention of batteries has had a drastic effect on the transportation industry as vehicle technology was able to overcome their major concern for fuel. The State of Charge (SOC) of a battery determines the charging and discharging rate of a battery, which in turn depicts its life expectancy. In the real world, the estimation of SOC in real-time is quite complex due to the internal parameters of the battery. In this project, deep learning algorithms, Real Driving Cycles (RDC), and IoT are used to predict the SOC of a vehicle. The Real Driving Cycle is constructed using the On Board Diagnostic II dongle for actual driving in a specified route, which is therefore collected as a .csv file using a diagnostic app. The deep learning algorithm predicts the speed in future time steps. The speed and throttle data with respect to time collected are then fed into the EV setup designed. As the motor of the EV setup operates, the corresponding battery consumption is recorded using a voltage sensor. In addition to that, speed is predicted in future time steps and respective battery consumption is recorded. Using battery voltage, SOC is calculated for both predicted and actual speed. The predicted data allows the user to foresee their journey concerning the charge of the battery. The concern of the assumption of the Indian Driving Cycle as India’s traffic congestion is homogeneous has been overcome by adopting RDC. As a result of experiments, it is confirmed that the bi-directional Long Short Term Memory model predicts SOC more accurately than the other Recurrent Neural Networks.