Power Electronics Converters Topology Derivation with Combination of TopoDiffVAE and Reinforcement Learning

With the widespread use and development of power electronics, the study of power electronics, especially power electronics topology, has become increasingly compelling. With the study and expansion of graph theory on power electronics topology, it has become an idea to describe and reveal the topology by graph structures. We propose a framework combining the TopoDiffVAE model and reinforcement learning to reveal and learn the intrinsic connection rules of a certain class of topology and generate new topology based on them. Our model can be used not only to generate new topology, but also to accelerate the generation of other neural network-based topology.