Surrogate models for power electronic systems applying machine learning techniques

As society advances toward a green transformation, digitalization, and Industry 4.0, power electronics have an important role in controlling and converting electric energy in a flexible, efficient, and reliable way. The increasing use of power electronics and the shortening of product development cycles are promoting design automation in the field of power electronics, which can reduce the complexity of the design process and provide excellent insight into optimization. A key element of implementing design automation is establishing efficient models. Surrogate models, which replace expensive numerical simulations and experiments by mathematical functions such as polynomial chaos expansions, Kriging, and neural networks, have become a promising method for engineering design and optimization. This chapter introduces the concept and workflow of surrogate modeling for power electronics. Based on typical applications of power electronics, surrogate modeling approaches in terms of active components, passive components, power electronics converters, and systems are demonstrated. Finally, an example of applying a surrogate model in the thermal modeling of power semiconductors considering cross-coupling effects is presented in detail. Through a surrogate model, quantities of interest can be quickly explored within sufficient accuracy while consuming less time and effort.