TU Delft’s team earned 2nd place in the innovation track of the 2023 MagNet Challenge, organized by Princeton University and the IEEE Power Electronics Society, awarded at APEC Applied Power Electronics Conference and Exposition 2024 in Long Beach, USA. This competition focused on using large datasets to develop new methods for accurately predicting power loss. For more details, check out the MagNet Challenge GitHub Repository: https://github.com/minjiechen/magnetchallenge

Our team, composed of Zhengzhao Li, Reza Mirzadarani, Ruijun Liu, Lu Wang, Tianming Luo, Dingsihao Lyu, Mohamad Ghaffarian Niasar, and Zian Qin, introduced a unique FFT+FNN technique that improves how we predict power loss. We made our model even better by using Multi-Objective Optimization (MOO) to make it both simpler and more accurate. We also used transfer learning to overcome challenges with not having enough data.

During the award ceremony, the competition organizer, Professor Minjie Chen from Princeton University, praised our Multi-Objective Optimization approach. He highlighted its significance for industrial applications, noting our method’s effective balance between model size and prediction accuracy.