Optimized Electric Vehicle Charging Using Real-Time Data and Machine Learning

This paper presents a software-defined, edge-cloud hybrid framework for intelligent electric vehicle (EV) charging station navigation, developed entirely without physical hardware. The system integrates four core components: a physics-based EV simulator, a Gradient Boosting range prediction model trained on real specifications from 103 commercially available EV models, a Random Forest demand forecasting module, and a dynamic queue management engine all deployed via a FastAPI backend with an interactive OpenStreetMap-based web dashboard. The range prediction model achieves a mean absolute error of 9.37 km and an R² score of 0.977, correctly differentiating vehicle-specific efficiency across models such as the Tesla Model 3 (161 Wh/km) and the Nissan Leaf (206 Wh/km), producing a 95.7 km range difference under identical driving conditions that naive linear formulas cannot capture. The demand forecaster predicts station occupancy with a training MAE of 0.031 occupancy rate units, capturing rush-hour, weekend, and temperature-driven demand patterns. A composite scoring mechanism ranks stations by distance, charger power, and real-time availability, while a multi-vehicle load-balancing algorithm with a 90% capacity penalty prevents pathological convergence of concurrent vehicles to a single station. The queue manager provides exact per-vehicle position and clock-time completion estimates, with a mean wait-time prediction error of 4.2 minutes across 50 test scenarios. The entire system is built on open-source tools (FastAPI, scikit-learn, Leaflet.js, OpenStreetMap) and requires no proprietary APIs, specialized hardware, or cloud subscriptions, making it fully reproducible on a standard laptop. This work serves as a practical, hardware-independent template for EV charging intelligence research and rapid prototyping.