Machine Learning-Based Approach for Modeling Power Consumption of Electric Vehicles

This paper introduces a data-driven, machine learning-based approach for estimating power consumption of an electric vehicle (EV) powertrain, using data obtained from MATLAB/Simulink. Most studies in this domain focus on presenting mathematical equations to model powertrain power consumption, often overlooking inherent non-linear behavior of powertrain components. These methods require many simplifying assumptions, accumulating errors in the final estimation of power and energy consumption. This work aims to address this challenge by developing a multilayer perceptron (MLP) to obtain a more accurate and less complex method to estimate power consumption of an EV powertrain, generalizable to any EV type. This machine learning-based model also accounts for a wide range of parameters, including battery state of charge (SOC), road grade, vehicle speed, and acceleration. Initially, power consumption data is generated using an EV MATLAB/Simulink model, then the MLP is trained on this dataset. The results indicate that the MLP model, when applied to the test set, exhibits a mean absolute percentage error of 0.11 %, showing that the model provides relatively reliable and consistent predictions and generalizes well across different power consumption ranges.