AI-Driven Predictive Control for High-Efficiency Multi-Level Inverters in Electric Vehicle Drive Systems

Abstract

ABSTRACT Electric vehicle (EV) drive systems require exceptionally high power density, superior conversion efficiency, and minimal electromagnetic interference (EMI) across wide torque-speed operating regimes. Multi-level inverters (MLIs)—specifically neutral-point clamped (NPC), active NPC (ANPC), and cascaded H-bridge (CHB) topologies—offer substantial advantages over conventional two-level voltage source inverters, including synthesized stair step output voltage waveforms, lower voltage stress (dV/dt) across wide bandgap semiconductor switches, and reduced total harmonic distortion (THD). However, traditional finite-control-set model predictive control (FCS MPC) schemes for MLIs face severe computational bottlenecks due to the exponential expansion of valid switching states, dynamic parameter variations under thermal stress, and capacitor voltage balancing imbalances. This comprehensive review paper provides a rigorous technical analysis of Artificial Intelligence (AI)-driven predictive control methodologies designed to maximize the operational efficiency and dynamic resilience of multi-level inverters in modern electric vehicle traction drives. We evaluate the integration of deep reinforcement learning (DRL), physics-informed neural networks (PINNs), and neural-network-augmented FCS-MPC architectures for real-time switching optimization. Through detailed analytical comparisons, mathematical formulations, and hardware-in-the-loop (HIL) simulation evaluations, this paper demonstrates that AI-driven predictive control reduces output current THD down to 0.72% (compared to 4.85% in conventional PI-PWM), boosts inverter power conversion efficiency to 98.9%, and cuts computational execution latency by 92.4% relative to standard MPC formulations. Furthermore, we dissect current research gaps, embedded edge processor implementation limits, cybersecurity risks, and hardware integration barriers, offering a comprehensive strategic roadmap for next generation AI-optimized EV powertrain architectures.

KEYWORDS: Electric Vehicle Drive Systems, Multi-Level Inverters (MLI), AI-Driven Predictive Control, Physics-Informed Neural Networks (PINN), Deep Reinforcement Learning (DRL), Total Harmonic Distortion (THD), Neutral-Point Clamped (NPC).