Developing a Data-Driven Energy Demand Predictor for Electric Vehicles in the UAE Using Statistical Analysis and Machine Learning

To address the pressing concerns surrounding electric vehicle (EV) range anxiety and energy availability, accurate prediction and coordination of the EV energy demand is vital. Whilst existing EV demand predictors rely on traffic simulators or local charging datasets, their effectiveness may be limited by a lack of actual EV driving data, hindering their scalability across different geographic regions. This study proposes a data-driven approach for EV demand prediction using both machine learning (ML) and statistical analysis, utilizing vehicular traffic flow data based on data collected on 250 EV trips across Dubai, UAE. The 250-trip data are used to obtain a trip-wise energy consumption model using SAS software, resulting in the simplified model consisting of the trip length variable, the trip duration variable and the average altitude of the trip variable. Furthermore, the data are then used to train various ML models including extreme gradient boosting (XGBoost) and random forests (RF), and deep learning (DL) models, such as multilayer perceptron (MLP) and long short-term memory (LSTM) to predict the trip-wise energy consumption. The performance metrics for both training and testing of various ML and DL algorithms reveal a stark contrast, as the models fit the training data exceptionally well but generalize poorly to unseen data. The decline in performance across all models from training to testing suggests issues such as over-fitting and inherent noise in the dataset, along with the relatively low data volume using 250 trips only. However, acquiring first-hand data for EVs in the UAE is a key contribution of this work, which lays the ground for further studies into UAE-specific EV electricity demand modeling.