Abstract: At present, the optimization research on wind buffeting noise of automobile mainly uses sound pressure level (SPL) as a single evaluation index, which cannot fully reflect the physical properties of noise, nor can it consider the subjective cognitive process of human ear to noise. The sound quality is introduced to evaluate the wind buffeting noise accurately. Firstly, the large eddy simulation (LES) is used to perform numerical simulation on the wind buffeting noise, whose accuracy can be judged according to the actual vehicle road test. Furthermore, on the basis of the numerical simulation results of wind buffeting noise, the objective and subjective evaluation of sound quality are carried out. The BP neural network prediction model of sound quality is established by integrating with sound quality objective parameters and subjective evaluation of sound quality. Finally, genetic algorithm (GA) is introduced to optimize the structural parameters of BP neural network, and a GA-BP prediction model of sound quality is established. The research results show that GA-BP sound quality prediction model is superior to BP neural network prediction model in training speed and prediction accuracy. The prediction model is based on the subjective and objective evaluation results of sound quality, and its predictive value can replace the traditional sound pressure level evaluation index and provide more accurate and reasonable evaluation for wind buffeting noise.

Key words: wind buffeting noise, sound quality, large eddy simulation, BP neural network, genetic algorithm