Machine-learning-based range estimation for electric vehicles with an Indian road perspective

Increased acceptance of electric vehicles is a good sign but the fear of range is one major challenge. In order to enhance their market penetration, better range prediction methods need to be devised. Most of the researchers working in this area have used public datasets, most of which are collected outside India. Being a developing country, with a wide range of ambient, weather and terrain conditions, electric vehicle range could vary significantly in different parts of India. Insufficient travel datasets on Indian roads are one major research gap that needs to be addressed. This research work uses actual driving data collected from a commercially available hatchback electric car model in India. Simulations are performed considering the appropriate vehicle specifications and ambient conditions. The effect of road grade, auxiliary load consumption and passenger count on the remaining range of the vehicle has been analysed and tabulated. Analysis is done on real travel data collected for three months. Mathematical equation-based energy consumption calculations and regression algorithm-based energy estimations are presented. Linear regression, multi-layer perceptron (MLP), gradient boost machine (GBM), support vector machine (SVM) and k nearest neighbour algorithms were implemented, and energy consumption prediction was done. The results were compared and tabulated.