Prediction Methodology for High-Frequency Sound Transmission Loss in Substation Acoustic Barriers Based on Bayesian Neural Networks

Conventional prediction and measurement methods for sound transmission loss (STL) in acoustic barriers face limitations in efficiently and accurately acquiring full-frequency-band STL data for substation acoustic barriers. To address this challenge, this study proposes a Bayesian Neural Network (BNN)-based prediction methodology for high-frequency STL in substation acoustic barriers. The method leverages low-frequency STL data obtained from in-situ measurements and key structural parameters (e.g., surface mass density) of the acoustic barriers to predict high-frequency STL, thereby enabling comprehensive full-frequency-band STL. Validation using an acoustic barrier dataset demonstrates the model's capability to accurately predict STL at 2 kHz and 4 kHz frequencies, confirming its effectiveness in addressing the targeted engineering problem.