AI-Based Surrogate Model for EMI Filter Optimization for Electric Vehicle Applications

Abstract:
Optimizing electromagnetic interference (EMI) filters for automotive inverters is a complex multi-objective problem, requiring a balance between efficiency, cost, and volume while ensuring compliance with stringent attenuation constraints. Traditional EMI filter design relies on iterative simulations and expert knowledge, leading to significant computational costs. This paper proposes a computationally efficient optimization workflow that replaces time-consuming numerical circuit simulations with a neural network (NN)-based surrogate model. The NN is trained using high-fidelity LTspice simulations and integrated into a genetic algorithm to accelerate the multi-objective optimization (MOO) process. The proposed approach is demonstrated using a two-stage differential-mode EMI filter for electric vehicle applications. The results show that the proposed approach achieves a massive reduction in computation time, enabling rapid exploration of the design space while maintaining a higher level of accuracy. This methodology is highly scalable and can be extended to full inverter optimization, significantly enhancing the feasibility of comprehensive power electronics MOO.