Realization of superhuman intelligence in microstrip filter design based on clustering-reinforcement learning.

Microstrip filters are widely used in signal processing because of their light weight, compact structure and high reliability. Designing these filters is very time consuming, and a designer generally needs much knowledge of electromagnetic theory. In recent years, artificial intelligence (AI) technology has been used to accelerate the design process. However, current AI models retain the human design mindset (adopt a regular structure) and thus cannot be applied to the automatic design of irregular structures. We proposed a clustering-reinforcement learning model named parallel advantage actor-critic with K-means (PAAC-K). The PAAC-K model is based on a reinforcement learning model, in which the size of the overlapping area is used as the reward function, and a clustering algorithm was added to extract characteristics for learning. We used the stepwise training method to avoid repeated exploration in a design with different frequencies. The PAAC-K model realized superhuman intelligence that automated the design of irregular structures, which was proven with four application examples. This work presents an AI model and a design mindset for irregular structures, which is of great significance in promoting the development of filter devices. [ABSTRACT FROM AUTHOR]