Optimal Prediction of EVs State of Charge based on Deep Machine Learning Technique

Abstract

Although Electric Vehicles (EVs) are widely used globally, they are still emerging in some markets, such as the Egyptian market. Consequently, there is a growing need to increase the adoption of EVs to harness their potential in addressing climate change and environmental challenges. Several studies have focused on EVs and the management of their charging and discharging processes to maximize the benefits of their integration into the electricity grid. In particular, current research emphasizes collecting EV data, creating datasets, and utilizing them for predictive purposes. Relying solely on historical data limits the potential to achieve accurate predictive results. Therefore, this proposed study places strong emphasis on integrating multiple features that directly influence prediction outcomes. A proposed novel technique was developed to predict the initial and required State of Charge (SoC). This technique combines historical charging data with critical real-world features, such as distance, road type, traffic patterns, and events data. The results of the proposed technique demonstrated clear superiority over existing studies through comparative evaluation metrics. This study achieved SMAPE scores of 0.0008 for the initial SoC and 0.00019 for the required SoC, outperforming prior literature. Through comparative analyses with previous studies using the same dataset, this research illustrates significant improvements in prediction accuracy, attributed to the integration of data on total distance, road type, traffic patterns, and weather conditions.