Sound-insulation prediction model and multi-parameter optimisation design of the composite floor of a high-speed train based on machine learning

Abstract

Designing a sound-insulation scheme for a composite structure efficiently and accurately for noise control in equipment is essential. However, traditional simulation and experimental methods for obtaining an optimal solution are not only time-consuming but also difficult to implement. In this paper, a sound insulation optimisation design method based on machine learning is proposed. The method is applied to design a complex composite floor structure of a high-speed train. By testing numerous practical schemes in the acoustics laboratory to obtain a sample set, a machine learning model for predicting the sound insulation performance of a composite floor of a high-speed train is trained and verified. Subsequently, an efficient and accurate multi-parameter sound insulation optimisation design of the composite floor structure based on the machine learning model is implemented. First, the original data samples required for model training are analysed and sorted. Second, the target feature subset is selected through the main influencing factor analysis, correlation-redundancy analysis, and mRMR feature selection calculation. Then, based on the SVR method, the standardised feature data are used to train and verify the sound-insulation prediction model of the composite floor structure of a high-speed train. Finally, two embodiments are presented to verify the advantages of the model in the multi-parameter optimisation design of the sound-insulation model of the composite floor structure of a high-speed train. The results show that the optimal sound insulation is 51.69 dB when the thickness and surface density of the composite floor are given. Similarly, the minimum surface density is 89.42 kg/m2 when the thickness and sound insulation limit are given.