Battery Consumption Modeling for Electric Vehicles Based on Artificial Neural Networks

This paper presents how to develop a battery consumption model taking advantage of state-of-charge streams acquired from real-life electric vehicles. From the record consisting of timestamp, longitude, latitude, and battery remaining, learning patterns are generated to build a neural network for each of 4 major roads, essentially taken by long-distance trips in Jeju city. Our 3-layer neural network model is made up of an input node, 10 hidden nodes, and an output node. The input variable takes the approximated distance while the output variable represents the battery consumption from the start point of a road. Neural networks, being able to efficiently tracing non-linear data streams, accurately keep track of battery consumption irrespective of road shapes and elevation changes. The assessment result shows that the average errors for each road range from 0.22 to 0.33 km, indicating that this model can estimate battery demand for a given route for navigation applications.