Research on Noise Prediction Methods for Sound Barriers Based on the Integration of Conditional Generative Adversarial Networks and Numerical Methods

Qian Hu ¹ Ziyu Cui ^2* Hongxue Liu ¹ Senhao Zhong ³

• ¹ Huanghuai University, Zhumadian, China
• ² Xinyang Normal University, Xinyang, China
• ³ University of South China, Hengyang, Hunan Province, China

This study presents a novel wideband acoustic state analysis approach utilizing Conditional Generative Adversarial Networks (CGANs). In conventional acoustic analysis using the Boundary Element Method (BEM), the frequency-dependent system matrix requires repeated computations across multiple frequencies, resulting in substantial computational costs. By expanding the Hankel function into a Taylor series, the frequency-dependent and frequency-independent components of the acoustic boundary integral equations are effectively decoupled. This decoupling enables the construction of a frequency-independent system matrix, thereby reducing computational complexity significantly. Furthermore, the asymmetric and full-rank nature of the BEM coefficient matrices exacerbate computational demands, particularly in large-scale problems requiring wideband frequency analysis with iterative equation solving. To address these limitations, a CGANbased modeling framework is proposed as an efficient alternative. This framework markedly reduces computational time while maintaining high predictive accuracy and exhibits exceptional adaptability across datasets with diverse characteristics and geometric configurations. Numerical experiments validate the proposed method, highlighting its superiority in both accuracy and computational efficiency.