基于RNR-WVD与GA-小波的非稳态排气噪声声品质研究

Study on the sound quality of unsteady exhaust noise based on RNR-WVD calculation and GA-wavelet network

基于心里声学客观参量的GA-BP声品质预测模型能够准确的预测稳态排气噪声声品质。对于非稳态噪声研究,引入正则化非稳态回归技术(RNR)优化计算维格纳-威尔分布(WVD)的时频方法,建立新的声品质参量SQP-RW(Sound Quality Parameter Base on RNR-WVD),用此参量替换掉与满意度相关性较小的客观参量。同时,以Morlet小波基函数作为隐含层结点的传递函数构建小波神经网络(Wavelet Neural Network,WNN),并用GA优化小波神经网络层间的权值和阈值,构造出GA-WNN并用于非稳态排气噪声声品质预测。结果表明:GA-WNN在非稳态排气噪声声品质预测上比GA-BP神经网络更加准确;引入SQP-RW参量,模型具有更高的精度,更能体现出非稳态信号特征及声品质特点。

The GA-BP acoustic quality prediction model based on the psychoacoustical objective parameters can accurately predict the sound quality of the steady state exhaust noise.For the research on unsteady state noise,RNR(regularization non-stationary regression technique)was applied to compute the WVD distribution(RNR-WVD),and a new parameter SQP-RW was built to replace the objective parameters that have less correlation with the satisfaction degree.Meanwhile,a WNN(Wavelet Neural Network,WNN)was constructed by using the Morlet wave basis function which was used as the transfer function of the hidden layer.And GA was used to optimize the weights and thresholds of the WNN layers,the GA-WNN was constructed and used to predict the acoustic quality of the unsteady exhaust noise.The results show that the GA-WNN is more accurate in predicting the sound quality of unsteady exhaust noise than the GA-BP neural network.With the introduction of SQP-RW,the model has higher accuracy and it can reflect the characteristics of unsteady signals and the sound quality characteristics.