Tagungsband - DAGA 2024
Beitrag
Double Deep Q-learning with Prioritized Experience Replay for Designing Noise-Mitigating Structures
Zusammenfassung:
The remarkable progress in deep reinforcement learning has enabled autonomous problem-solving in complex tasks, yielding unprecedented results across various fields. However, the application of these techniques in acoustics is still in its early stages. In this paper, we investigate the potential of these algorithms to design optimal structures that mitigate noise in transmission and reflection problems. Specifically, we utilize the double deep Q-learning algorithm, with convolutional neural networks and prioritized experience replay, to explore various designs aimed at maximizing the objective function, i.e., absorption bandwidth. Additionally, we compare its performance with heuristic approaches, such as the genetic algorithm, and fully connected layers. We provide a detail discussion of the parameters that influence the algorithm’s performance. Remarkably, even without data labeling or prior knowledge of the environment, the algorithm quickly learns to design noise-mitigating structures.