Predicting environmental noise using neural networks

摘要

Advancements in acoustic instrumentation and data acquisition enables the analysis of a substantial amount of data collected during long term noise monitoring programmes. Generally, variations of environmental noise at a particular location are random. However, recent research and analysis based on a substantial amount of data collected in urban areas showed the presence of patterns or major cycles in the time histories of noise levels. This paves a way to exploring the concept of a generic time history and predictions of future noise. Neural networks represent an attractive tool to explore and predict future noise levels based on previous measured values. This paper considers opportunities for predicting noise levels using this tool. The networks were trained on noise data acquired in a variety of urban and suburban noise environments, such as road, rail and industrial noise sources. Different modelling approaches are considered for attempts to forecast a noise level time history. Future use of this technique in environmental noise monitoring could open opportunities for expecting or preventing non - compliance.