Study On The Prediction Of Car Acceleration Noise Sound Quality Based On Support Vector Machine

Because the interior noise of the car process a large number of low-frequency components in the running, in the past to the evaluation criteria of the A-weighted sound pressure level noise can not fully reflect people’s subjective feelings. Therefore, the research on the interior noise sound quality has important practical significance.Firstly, according to GB/T 18697-2002, the experiment was designed on the acquisition of car interior accelerating noise. 14 different brands of cars interior noise signals were collected in 30 ~ 80 km / h acceleration time by using the B&K PULSE system, and the collecting signals were pretreated. 14 non-loudness noise samples were got and noise loudness, roughness, sharpness, volatility and A-weighted sound pressure level of each sample were calculated. The subjective evaluation test had been used to get each sample subjective evaluation value on the 14 noise samples by using the paired comparison, and the objective parameters having significant effect on the sound quality were extracted by the correlation analysis, using objective parameters as the input and subjective parameters as the output, A car interior accelerating sound quality prediction model was built based on support vector machine, and being verified by the test samples.Secondly, in order to exclude loudness factor, a sample was equal-loudness processed. 10 noise samples were obtained and roughness, sharpness, volatility and A-weighted sound pressure level of each noise sample were calculated. The subjective evaluation test has been used to get each sample subjective evaluation value on the 10 noise samples by using the paired comparison, and the objective parameters having significant effect on the sound quality were extracted by the correlation analysis, using objective parameters as the input and subjective parameters as the output, A car interior accelerating sound quality prediction model was built based on support vector machine, and being verified by the test samples.Finally, compared the relative error of the support vector machine prediction model with multiple linear regression model, SVM prediction relative error is significantly lower than the multiple linear regression model verifying the superiority of SVM model. Also 7 equal-loudness models are compared, the results show: under circumstances of the equal-loudness, roughness and A sound level are major factors on the accelerating condition interior noise sound quality.In short, based on support vector machine,the car interior accelerating noise sound quality models and researching method can accurately evaluate and predict sound quality, research can improve the sound quality of the car, thereby it is so important to increase automotive NVH control level.