High-Precision Modeling and Prediction of Acoustic Comfort for Electric Bus Based on BPNN and XGBoost

Abstract

At present, the A-weighted sound pressure level inside electric buses has generally reached the industry decibel limit, and sound quality research is a considerable way to improve future vehicle performance. In this paper, 64 noise samples from eight electric buses are collected, with acoustic comfort as the evaluation index, the subjective evaluation tests are carried out by rank score comparison (RSC), and nine objective psycho-acoustic parameters of all the samples are calculated to form a basic database. Aiming at the high-precision modeling requirement of electric bus sound quality and taking objective parameters and acoustic comfort as input and output variables, two machine learning algorithms, back propagation neural network (BPNN) and extreme gradient boosting (XGBoost), are respectively performed to establish nonlinear comfort evaluation models through data training, and ultimately, based on sample data test and relative error comparison, the acoustic comfort evaluation model with prediction accuracy of 95.65% and its mathematical formula are determined. This lays a key technical foundation for the future evaluation and optimization of electric bus sound quality.