Reinforcement learning applied to an Active Noise Control system in Sonic Crystal Noise Barriers

Abstract

Noise control is one of the main environmental challenges today. Transport noise is often reduced by using noise barriers which are problematic due to their lack of permeability to water and wind. In recent decades, an alternative to conventional noise barriers has been proposed based on Sonic Crystals, called Sonic Crystal Noise Barriers (SCNB). However, they present another problem due to their lack of efficiency at low frequencies. On the other hand, cheaper technology in recent years has greatly enhanced Active Noise Control (ANC). Protection against noise is becoming more and more feasible using this type of technology. In contrast to SCNB, active noise control is more efficient at low frequencies because the sweet point is larger at these frequencies. All this leads us to consider the combination of both ideas in this numerical approach. The aim is using a Reinforcement Learning (RL) architecture with a Double Deep Q-Network (DDQN) agent, studying the potential of incorporating ANC into these SCNB and produce permeable barriers with the best low frequency response possible.