基于神经网络的车辆排气噪声声音品质预测技术

Prediction of Vehicle Exhaust Noise Based on Neural Network

摘要:

通过评审团成对比较法测试得到18种车辆排气噪声的满意度评价，考察并选取响度、尖锐度、粗糙度、波动度和峭度作为描述车辆排气噪声声音品质的客观心理声学参数，使用BP神经网络理论建立车辆排气噪声声音品质神经网络预测模型，对排气噪声样本的满意度进行预测，并与使用多元线性回归模型所得的预测值进行了比较。结果表明，神经网络模型预测值更接近实测值，误差在10%范围以内，对于单一噪声样本满意度的预测精度高于多元线性回归模型，能够更好地反映客观参数和主观满意度间的非线性关系，可用于车辆排气噪声声音品质的预测研究。

关键词:车辆;声音品质;排气噪声;神经网络;预测模型

Abstract:

Sensory pleasantness evaluation of eighteen vehicle exhaust noises were obtained by paired comparison jury test. Loudness, sharpness, roughness, fluctuation strength and kurtosis were selected for objectively characterizing the sound quality of exhaust noise. The sound quality prediction model of vehicle exhaust noise was established based on back-propagation neural network. Sensory pleasantness of exhaust noise samples were obtained through the prediction model and the results were compared with that obtained through multiple linear regression prediction model. The result showed that the prediction values were close to the measured values, the neural network model was more effective than multiple linear regression model in prediction of individual exhaust noise. The neural network prediction model represented the nonlinear relation between sensory pleasantness and objective parameters exactly and could be used for predicting the sound quality of vehicle exhaust noise.

Key words:车辆;声音品质;排气噪声;神经网络;预测模型