An online prediction method for ship underwater radiated noise based on differential evolution feature optimization and ensemble method
Xin Huang, Rongwu Xu, Ruibiao Li
Abstract
Accurate real-time prediction of ship underwater radiated noise (URN) is significant for ship URN monitoring and reduction. To tackle the complex vibration-acoustic transfer mechanisms, the limited vibration monitoring data, and the scarcity of navigational trial samples, this study proposes an online URN prediction model that integrates dual-dimensional feature optimization and an improved ensemble method (EM). The framework consists of three key innovations. Firstly, a dual-dimensional optimization strategy based on differential evolution (DE) is introduced, which simultaneously optimizes sensors and 1/3-octave frequency bands by minimizing training and validation errors, enhancing data utilization by over 40 %. Secondly, an improved Bagging EM model with Ridge regression as base learners is proposed to integrate multi-source features effectively. Finally, the vibration-acoustic radiation experiment on a cabin model demonstrates that the proposed method achieves prediction errors below 3 dB in 97 % of the frequency bands under test conditions, reducing the maximum error from 11.97 dB to 3.42 dB, which indicates its effectiveness in predicting ship URN.