Development of an AI-based Energy Prediction for Electric Vehicles

Abstract

The rise of electric vehicles offers new challenges as increasing energy efficiency and electric range through energy management systems as well as avoiding range anxiety by providing reliable range information. To realise this, predictive approaches based on a calculation of the expected energy consumption for the route to be driven are a feasible option. However, such approaches, realised through appropriate algorithms, usually require significant computing time, which can hinder their application. Therefore, this paper presents an approach to realise energy predictions using artificial intelligence (AI) approaches. The training of these AI models is performed with simulated data, generated by the algorithm to be replaced. Three AI models are build, trained, evaluated and optimised to predict a vehicle’s energy consumption. A feed forward (FNN ) and a recurrent neural network (RNN ) model utilise deep learning approaches while a XGBoost model represents conventional machine learning techniques. In conclusion, the deep learning models struggle to match the results of the reference prediction algorithm, while the RNN model even fails to reduce calculation times. In contrast, the XGBoost model is able to generates accurate energy predictions, while drastically reducing the calculation time.