Predictive Modeling of Electric Vehicle Charging Duration Using Multilayer Perceptron Neural Networks

Abstract

Accurate prediction of electric vehicle (EV) charging duration is essential for efficient energy management and enhanced user experience. However, variability in charging patterns, battery conditions, and operational factors makes reliable prediction challenging. While charging duration is inherently continuous, practical EV charging operations often require approximate duration ranges rather than exact times. Therefore, this study reformulates the prediction task as a classification problem by discretizing charging duration into predefined categories.
A Multilayer Perceptron Neural Network (MLPN) is employed to classify EV charging durations as Low or Long. The model is trained on a real-world dataset containing 2,000 charging sessions, with relevant features such as energy consumed and temporal attributes derived from charging start times. Preprocessing steps, including normalization and feature selection, are applied to enhance model accuracy.
Two optimization algorithms, Stochastic Gradient Descent (SGD) and Adaptive Moment Estimation (Adam) are evaluated for their impact on model performance. Classification metrics including accuracy, precision, recall, and F1-score are used for evaluation. The results indicate that the MLPN model accurately classifies charging duration, with SGD achieving superior performance. The proposed approach provides a practical data-driven solution for EV charging duration prediction, supporting efficient energy utilization and improved operational planning.