The growing adoption of electric vehicles (EVs) requires innovative energy management strategies to stabilize off-grid charging stations, balance energy demand, and reduce power losses. Traditional energy management systems (EMSs) face challenges maintaining efficient, cost-effective operation, especially during peak demand periods. This study introduces an AI-based scheduling approach, integrating machine learning (ML) models with a Grey Wolf Optimizer (GWO) for optimized EV load demand forecasting and scheduling. Our regression model achieves a load prediction accuracy of 93%, facilitating adaptive scheduling that reduces grid losses by managing load variability. Experimental results show that the proposed system enhances energy efficiency by up to 15%, minimizes power losses by up to 262 kW, and stabilizes energy demand fluctuations from peak values of 450 kW to optimized levels. Developed and tested in MATLAB, this AI-driven EMS significantly reduces operational costs and enhances reliability in off-grid EV charging, contributing to sustainable and efficient energy utilization.