The paper discusses the challenges in power electronics parameter design, usually addressed using detailed optimization approaches or brute force grid search. It introduces a new method, Continuously Adapting Random Sampling (CARS), which provides a continuous method, allowing for fast and large amounts of simulations while focusing on the most promising parameter ranges. The method draws inspiration from multi-armed bandit research and results in prioritized sampling of sub-domains in one high-dimensional parameter tensor. The performance of CARS was evaluated on three power electronic use-cases and found to be competitive with genetic algorithms, and additionally allows for highly parallelizable simulation.
Publication date: 17 Oct 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2310.10425
