A Comprehensive Leakage-Free Forecasting Pipeline for Segmented Time Series: Application to Cross-Trip State-of-Charge Prediction in Automated Electric Vehicles

The rapid adoption of Electric Vehicles (EVs) in the global pursuit of energy efficiency and carbon neutrality necessitates effective strategies to mitigate their carbon footprint and enhance operational stability. Similarly, in order to achieve Sustainability Development Goals, a promising solution toward green mobility, which is gaining ground nowadays, constitutes Automated Vehicles (AVs), which are EVs having the capability to move autonomously, without the need for a driver. One of the most critical factors regarding energy efficiency is the optimal management of energy consumption of AVs. This research study explores the application of machine learning (ML) models for State-of-Charge (SoC) forecasting in AVs, crucial for addressing challenges such as range anxiety and grid overloading. Leveraging real-life EV data from automated minibuses in Gothenburg, Sweeden, a comprehensive pipeline is proposed for data pre-processing, feature selection, and model training. With a focus on predicting SoC several minutes ahead, various ML techniques, including linear regression, ridge regression, lasso regression, and elastic-net regression are embedded in a pipeline specifically developed to overcome the challenge of training time-series models on discontinuous data segments, corresponding to discharge cycles. This pipeline is called Cross-Segment-Leakage-Free (CSLF). The results demonstrate the efficacy of CSLF, with the best-performing model achieving a Mean Absolute Error (MAE) of 0.92 in a forecasting horizon of 30 minutes, representing a significant improvement over baseline models. The study underscores the importance of meaningful pre-processing and model selection in SoC consumption forecasting for AVs, offering insights into future research directions and deployment strategies for enhancing EV efficiency and grid stability.