Predicting Electric Vehicle Energy Consumption from Field Data Using Machine Learning

Licentiate thesis, 2024

This study addresses the challenge of accurately forecasting the energy consumption of electric vehicles (EVs), which is crucial for reducing range anxiety and advancing strategies for charging and energy optimization. Despite the limitations of current forecasting methods, including empirical, physics-based, and data-driven models, this paper presents a novel machine learning-based prediction framework. It integrates physics-informed features and combines offline global models with vehicle-specific online adaptation to enhance prediction accuracy and assess uncertainties. Our framework is tested extensively on data from a real-world fleet of EVs. The best global model, quantile regression neural network (QRNN), demonstrates an average error of 6.30%, the online adaptation further achieves a notable reduction to 5.04%, and both surpass the performance of existing models significantly. Moreover, for a 95% prediction interval, the online adapted QRNN improves coverage probability to 91.3% and reduces the average width of prediction intervals to 0.51 kWh. These results demonstrate the effectiveness and efficiency of utilizing physics-based features and vehicle-based online adaptation for predicting EV energy consumption.