Data-Driven Digital Twin for EV Energy Consumption Prediction

Abstract

Driven by environmental requirements and new technologies, electric vehicles (EVs) are rapidly gaining momentum due to their environmental and economic advantages. EVs are gradually replacing fuel-powered cars to become main mode of transportation. However, a major obstacle to the widespread adoption of EVs is range anxiety. Drivers worry that their vehicles do not have enough range to reach their destinations. To alleviate range anxiety, it is essential to accurately predict the energy consumption of EVs between the departure and the destination. This Master Thesis proposes data-driven methods for predicting EVs energy consumption. The data used included vehicle data provided by Zeekr Technology Europe AB and environmental data provided by external APIs. Vehicle and environmental data were processed into distance sequence data. When selecting energy prediction parameters, it was found that speed is required as a feature for energy prediction. Therefore, energy prediction was divided into two parts: speed prediction and energy prediction. Long short term memory (LSTM) model was used for speed prediction. For predicting energy consumption, a vehicle dynamics model was built as baseline. Basic machine learning models (linear regression, decision tree, random forest, and knearest neighbor.) and convolutional neural network with long short term memory (CNN-LSTM) model were used to predict energy consumption. The results of speed prediction were better than those of the map API and the existing model. The conclusion of the energy prediction was that all machine learning models performed better than the vehicle dynamics model. For short distances, the decision tree model provided the best predictions, while for long distances, the CNN-LSTM model offered the best predictions.