Machine Learning-Enabled Power Converter Control for Efficient Electric Vehicle Energy Management

Abstract

This paper proposes Machine Learning (ML)-based power converter control as an effective Electric Vehicle (EV) energy management strategy under time-varying source, load and storage conditions. The system contains a Hybrid Energy Storage System (HESS) unit, Photovoltaic (PV) source, grid input, bidirectional Direct Current-to-Direct Current (DC-DC) converter, DC-link bus, and traction inverter together with an EV motor/load and ML controller. The ML controller utilizes voltage, current, state of charge (SOC), temperature and load demand feedback to predict adaptive duty-ratio command values for stable energy delivery by the converters. The framework tackles the public utilization of EVs with Vehicle-to-Grid (V2G) and Grid-to-Vehicle (G2V) operation based on the mentioned elements like enhancing source selection, load prediction, converter efficiency, and battery protection. Simulation results reveal that the ML-enabled control achieves converter efficiency, DC-link voltage stability, and power loss reductions of 92.4-96.8%, 89.6-97.2%, and 68.5-86.4%, respectively, while Total Harmonic Distortion (THD) reduction increases from 70.2 to 88.7%. The simulation results of power-flow control are also analyzed, and show better PV utilization efficiency of PV units, less dependency on grid supply and generation, less deviation of SOC, and more regenerative braking energy recovery with a longer battery life. As a result, the ML-enabled converter control framework presents an adaptive, reliable and energy-efficient technology for solving complex power management problems encountered in next-generation EV applications.