Development Of Construction Noise Prediction Method Using Deep Learning Model

Siew, Jun Teng (2021) Development Of Construction Noise Prediction Method Using Deep Learning Model. Final Year Project, UTAR.

PDF Restricted to Repository staff only until 31 December 2023. Download (3679Kb)

Abstract

Current noise mitigation strategies are ineffective by only relying on passive and restricted noise management. The best practice to minimise noise pollution is to take precautions against noise hazards through noise assessment in the preconstruction stage. Despite that, a few case studies can be found for noise prediction, which has a lack of study in construction noise prediction. A simple prediction chart method was developed on top of a stochastic algorithm called Monte Carlo simulation by complying with the standard BS 5228 for the noise prediction in the environmental impact assessment during the planning stage of a construction project. In this study, a deep learning-based noise prediction model was proposed to improve this method. This model could relieve the burden of manual calculations for simple prediction charts, and no previous study has applied this algorithm to construction noise prediction. Based on the concept of a simple prediction chart, several ways for improvement, such as the duty cycle of the machinery and the receiver angle of 360°, can be included in the training noise data. In this project, thousands of deep learning models were trained and evaluated to select the best performance models for establishing a noise prediction model. Seven deep learning models trained by seven noise datasets with different aspect ratios were selected and implemented in the proposed noise prediction model. The mean absolute errors (0.2 to 0.25), root mean square errors (0.3 to 0.42), and explained variance scores (0.9975 to 0.9988) of the selected best deep learning models indicate an incredibly accurate, fewer outliers, and extraordinarily reliable noise prediction model. Such performance could be achieved by relying on several input data, including the duty cycles of machines, the locations of workers, and the dimension of the construction site. These data could be obtained easily during the planning stage of a construction project. The performance of this model is believed to be superior to the majority of manual predictions using a simple prediction chart. It will eventually become mainstream of the construction noise prediction method and will also be used in industries other than construction.

Actions (login required)