An efficient acoustic energy harvester by using deep learning-based traffic prediction

With rapid urbanisation, traffic noise pollution has become a growing community concern. However, traffic noise can be valuable due to its ubiquity and cleanliness, which can be exploited for acoustic energy harvesting systems. This paper aims to develop an efficient acoustic energy harvester to improve energy utilisation efficiency and reduce noise levels by using deep-learning methods. We propose a temporal graph attention convolutional network (TGACN) to predict noise frequency, providing guidance for the acoustic energy harvester design. Power efficiency can be maximised by adjusting the harvester's neck size to the frequency predicted by the TGACN model. The relationship between neck size and the optimum frequency range is obtained through simulations and experiments. We analysed the performance differences among the various cavities in multiple-cavity acoustic energy harvesters and discussed the circuit connection methods between each cavity to identify the optimal configurations. Additionally, we evaluated the superiority of TGACN over the state-of-the-art methods in terms of prediction accuracy by conducting extensive experiments with two real-world datasets.