Currently, the demand for power electronic energy conversion in various applications is continuously growing, making it challenging for conventional power electronic topologies to meet the increasingly diverse and stringent performance requirements, and there is a trend towards the customization of power electronic topologies. This paper conducts a comprehensive survey over the field of power electronics topology derivation, which elaborates on methods including empirical topology generation based on expert knowledge, enumerative topology generation through computer programming, and self-learning topology generation using deep reinforcement learning, analyzing the strengths and weaknesses of each approach. Furthermore, the paper explores cutting-edge developments in the field, proposing a novel approach that integrates large language models with deep reinforcement learning. By leveraging the language comprehension and code generation capabilities of large language models, this approach aims to provide a higher-quality design starting point for deep reinforcement learning, which demonstrates the potential in enhancing the efficiency of topology derivation.