Research On Prediction Of Vehicle Inner Sound Quality Based On Wavelet Neural Network

With the development of society,people’s requirements for NVH performance of automobiles have been gradually improved.The sound quality of interior automobile noise affects the overall NVH performance.At present,pure electric vehicles(EVs)are developing rapidly.Compared with fuel vehicles,the overall noise level of EVs is lower.However,the motor noise because of lacking noise masking such as engine is more unacceptable.Therefore,it is of great practical significance to study the sound quality of EVs.This paper is devoted to finding a more accurate and adaptable subjective and objective sound quality evaluation model.The pure electric vehicle is selected as the research object in this paper.The sound quality evaluation and prediction model of pure electric vehicle interior noise is analyzed and studied.The main research contents of this paper include:1)Firstly,this paper chooses three kinds of pure electric vehicles.Then noise acquisition equipment were used to acquire their uniform speed noise signals under different working conditions.After screening,intercepting and equal loudness processing of the original noise samples,a noise signal sample bank with 27 valid samples was established.Five traditional psychoacoustic parameters of noise samples were calculated.Next the sound quality of noise samples was objectively evaluated by grading method.The correlation between these five parameters and subjective score was analyzed.Finally three objective parameters with higher correlation were selected.2)The samples of noise signal sample bank are filtered and re-sampled.The difference of fractal dimension in time-frequency domain and the energy ratio coefficient of each node after decomposition of EEMD are extracted as new characteristic parameters of noise samples based on fractal dimension and wavelet packet decomposition theory.3)The time-frequency domain fractal dimension difference,wavelet packet energy ratio coefficient and three traditional psychoacoustic objective parameters based on EEMD decomposition reconstruction are taken as input parameters respectively.Then the subjective evaluation results of noise samples are taken as output parameters.A PSO-WNN prediction model based on adaptive mutation operator optimization is constructed,which is the same as that based on traditional wavelet neural network.Finally the prediction models are compared.The results show that the prediction effect of PSO-WNN based on adaptive mutation operator is the best when the fractal dimension difference in time-frequency domain based on EEMD is used as input parameter.At the same time,through comparative analysis,it is found that the accuracy and stability of the optimized WNN prediction model are better than that of the traditional WNN prediction model.