Deep Reinforcement Learning Based Optimization of Microwave Microfluidic Sensor

Abstract

The resonant structure of microwave microfluidic sensors is crucial to their performance. However, traditional manual design methods rely heavily on expert experience and extensive parameter tuning, making it difficult to achieve optimal performance. Thus, there is an urgent need for an automatic design method for resonant structures. This letter proposes a topology optimization method based on deep reinforcement learning (DRL) to optimize the resonant cavity structure within the sensor. The optimization algorithm uses a reward strategy to obtain the optimal structure, increasing the relative frequency shift of the sensor from 0.4 to 0.658, thereby enhancing sensitivity by 64.5%. Experimental results demonstrate that this method can effectively improve the sensitivity of microwave microfluidic sensors and exhibit robustness and versatility.