Prediction and optimization of acoustic absorption performance of quasi-Helmholtz acoustic metamaterials based on LightGBM algorithm

Optimization of acoustic metamaterial structures and prediction of acoustic absorption properties have received much attention in various fields. This thesis aims to optimize and predict the acoustic absorption performance of quasi-Helmholtz acoustic metamaterials by using LightGBM algorithm in machine learning. In this study, the hole diameter, hole neck length, pore wall thickness, cavity wall thickness, and inner cavity depth of the quasi-Helmholtz acoustic metamaterial are selected as the characteristic factors, and the peak absorption coefficient of the quasi-Helmholtz acoustic metamaterial tended to be close to 1 and the minimum frequency of the peak sound absorption was selected as the optimization objectives. The data of sound absorption coefficients and frequencies under different structural parameters were obtained through COMSOL acoustic simulation, and the corresponding datasets were constructed accordingly. The data were then trained using the LightGBM model, and the prediction model was evaluated to demonstrate the reliability and accuracy of the adopted machine learning method. The results were interpreted using the Shapley additive explanations model to explore the potential relationship between the characterized factors and the target variables. In this paper, by analyzing the model interpretation and optimization results, the optimal values of each parameter are obtained to satisfy the optimization requirements. The results of this study show that the acoustic performance of quasi-Helmholtz acoustic metamaterials can be predicted and optimized using machine learning methods. The study in this paper combines the method of machine learning with acoustic problems to provide a fast method for predicting the absorption performance of acoustic metamaterials.