Enhancing urban EV integration: a data-driven hybrid approach to charging station optimization and energy management

Abstract

Electric vehicles (EVs) are pivotal to sustainable urban mobility, but their large-scale adoption in developing cities depends on efficient charging infrastructure and grid stability. This study proposes a hybrid deep learning framework to optimize EV charging station placement and energy scheduling in Vijayawada, India, projected to host 70,000 EVs by 2028. A convolutional neural network (CNN) is employed to classify charger types (Fast vs. Level 2) based on spatial features such as geospatial coordinates, population density, and traffic volume, while a long short-term memory (LSTM) network forecasts hourly charging demand using synthetic 24-hour sequences. The dataset comprises 108 candidate locations, designed to mirror real usage patterns. Model performance is evaluated using classification accuracy and mean absolute error (MAE). Results indicate that the CNN achieved 92% accuracy in charger type prediction, while the LSTM produced an hourly demand forecast with an MAE of 25 sessions/hour. These outcomes demonstrate the framework’s ability to reduce grid stress by shifting peak loads and strategically placing chargers in high-demand zones. The study provides a scalable and adaptable solution for EV infrastructure planning, enabling resilient grid integration, and supporting sustainable urban energy systems.