Prediction of the sound absorption performance for micro-perforated panel based on machine learning

Abstract The sound absorption coefficient is a crucial indicator of the sound absorption performance of micro-perforated panels. Achieving data-driven predictions for micro-perforated panels is of significant importance for selecting panels with ideal sound absorption properties. In the paper, the dataset is established for the structure and sound absorption coefficient of perforated panels. The structural parameters of perforated panels include plate thickness, cavity depth, and perforation ratio. Each sound absorption coefficient curve needs to be discretized during dataset construction. Six feature frequencies, including 100, 200, 300, 400, 500, and 600 Hz, are taken. The importance index across the six frequencies is analyzed by the RF model. Subsequently, GA-SVR, PSO-SVR, CNN, and RF four models are employed to conduct comparative predictive analysis on the 2000 dataset. The prediction performance, stability, and generalization ability of the four predictive models are evaluated using R 2 , MAE, and RMSE. The results indicate the GA-SVR model demonstrates the best predictive capability across all frequencies on both the training and test sets.