基于降维-支持向量回归的车内稳态声品质预测  

The prediction of car interior steady sound quality based on dimension reduction-support vector regression

作　　者：夏小均 赖诗洋 徐中明[3] Xia Xiaojun;Lai Shiyang;Xu Zhongming(National Bus Quality Supervision ＆ Inspection Center,Chongqing Vehicle Test ＆ Research Institute Co.Ltd.,Chongqing 401122,China;College of Mechanical Engineering,Chongqing Vocational Institute of Engineering,Chongqing 402260,China;College of Automotive Engineering,Chongqing University,Chongqing 400030,China)

机构地区：[1]重庆车辆检测研究院有限公司国家客车质量监督检验中心,重庆401122 [2]重庆工程职业技术学院机械工程学院,重庆402260 [3]重庆大学汽车工程学院,重庆400030

出　　处：《现代制造工程》2018年第11期1-6,28,共7页Modern Manufacturing Engineering

基　　金：国家自然科学基金项目(51275540);重庆工程职业技术学院院级科研项目(KJB201712)

摘　　要：基于信号分析与机器学习方法,提出基于降维-支持向量回归(Dimension Reduction-Support Vector Regression,DRSVR)的声品质主观预测模型。以车内稳态声样本为研究对象,计算并分析了其基本物理参数、心理声学参数。运用成对比较法对声样本进行了主观偏好性实验,验证了烦恼度(Psychoacoustic Annoyance,PA)模型初步判断该类样本声品质好坏的可用性。通过因子分析、聚类分析与相关分析,完成了声样本的降维,提取出了主要影响参量,再以支持向量回归的方法建立了主观评价预测模型。相关分析显示,基于降维-支持向量回归预测模型的计算值与主观评价值的相关性较高,其预测能力比未降维的支持向量回归模型更优,证明运用DR-SVR方法对车内稳态声品质预测是有效的。The predictive model of Dimension Reduction-Support Vector Regression （DR-SVR） method to evaluating sound quali- ty of car interior with small samples was proposed. The interior stable sound samples were selected as the object to analyze and its physical parameters and psychoacoustics parameters were calculated. Subjective testing was carried out via the paired comparison method,indicating that it is available to justice the merits of such samples preliminary with Psychoacoustic Annoyance （PA）. Dimension reduction was achieved through factor analysis, cluster analysis and correlation analysis, then the predict model was established base on SVR method, and the correlation coefficient between predictive value and evaluated value of DR-SVR model was higher than the model without dimension reduction, verifying the method to predict subjective preferences of car interior sound with DR-SVR was effective.

关 键 词：支持向量回归 降维 声品质 主客观评价 预测 

分 类 号：U461.4[机械工程—车辆工程]