LSTM-based Prediction of EV Charging Behavior Using Deep Learning

Electric vehicles (EVs) have become central to the evolution of smart cities, especially for reducing carbon emissions and contributing to cleaner urban environments. However, as more EVs are adopted, the increased demand on the electrical grid presents a significant challenge. Optimizing public charging infrastructure through advanced scheduling approaches is essential to handle the growing demand effectively. One promising approach involves integrating machine learning algorithms with scheduling systems to better anticipate EV charging requirements. By leveraging historical data on charging sessions, machine learning models can predict factors such as anticipated departure times and specific energy needs for each session. While previous studies have largely focused on generic charging demand forecasts, there has been limited investigation into how external factors—such as weather patterns, traffic flow, and local events—affect these predictions. Addressing this gap, this study explores variations in user behavior across a dataset of 267 electric vehicles with 59,520 charging events, aiming to enhance the accuracy of residential charging predictions. This approach not only aligns with peak grid capacity but also improves user satisfaction by ensuring reliable and efficient energy distribution. Our study also introduces an LSTM-based deep learning model that combines past charging records with additional contextual data, including weather, traffic, and event-based information, to more accurately forecast charging session durations and energy consumption needs.