Topology Derivation of Multiport DC–DC Converters Based on Reinforcement Learning

Abstract

Multiport dc–dc convertersare attracting wide attention in various applications. However, conventional topology derivation methods for multiport dc–dc converters are usually intricate and time-consuming. In this article, a reinforcement learning (RL)-based topology derivation method is proposed, which can derive topologies of complex converters quickly. To apply the RL framework, the topology derivation process is regarded as a Markov decision process. In each step, the agent selects and connects two components until a complete topology is made. To ensure that the derived topologies are feasible and match given voltage specifications, basic circuit constraints and duty cycle constraints are added as hard constraints. Soft constraints are also added to obtain optimal circuits with low voltage stresses or low current stresses. The testing on topology derivation of multiport dc–dc converters shows the great speed of the proposed method. Simulation and experimental results verify that the derived topologies cannot only satisfy given voltage specifications, but also achieve optimization targets of low voltage stresses or current stresses.