Noise Prediction on Offshore Support Vessels Using Artificial Neural Networks

Offshore support vessels, such as PSV and AHTS, are equipped with main and auxiliary engines with high installed power when compared to their size. Therefore, we note that the sound levels measured on these vessels are considered high and as a way of protecting the health of the crew and marine fauna, they are governed by IMO and classification society legislation. To date, noise on ships and offshore platforms is estimated using semi-empirical and/or statistical methods that present good accuracy in some areas, but significant discrepancies in others. Machine Learning (ML) techniques applied to Databases (DB) show promise, enabling accurate estimation of total noise on ships. In this work, artificial neural networks (ANN) were used to estimate sound pressure levels (NPS) in navigation (transit) and dynamic positioning (DP) conditions, in the compartments of five AHTS sister ships. The model selection criterion was an assessment of the effectiveness of the algorithms using indicators such as R2 and Mean Square Error (MSE). Finally, the models achieved MSE of 5.83 and R2 of 93.5% in DP, 6.71 and 91.2% in Transit and after using SMOTE, which is a data augmentation technique, they achieved MSE of 5.77 and R2 of 94.2% in DP and 13.51 and 95.5% in Transit.