摘要
为了解决旋转机械故障模式识别中传统时频分析法误识率过高问题,采用交叉验证和K近邻原理,提出了交叉验证K近邻算法.对标准测试数据集UCI进行了识别,分析了误识率和拒识率.仿真结果表明:当测试数据集取为UCI时,在规定的K值取值区间内取不同K值时,拒识率均为0;在规定的K值取值区间内,K值愈小,近邻数愈少,不能包含有用训练样本,导致误识率增加;K值过大,噪声显著影响误识率;通过交叉验证K近邻算法得到最优K值,使模式识别误识率达到最低.
In order to solve the problem of high incorrectness rate caused by traditional time-frequency analysis used in pattern recognition of rotating machinery fault, cross validation and K-nearest neighbor principle is used. Validation K-nearest neighbor algorithm is proposed to identify the standard test data set UCI and analyze rejection rate and incorrectness rate. The results show： While the test data set is UCI,taking different K value in the specified K value interval,the rejection rate is 0; While K value is small,the number of the nearest neighbors is less, which cannot contains useful training samples and results in increased incorrectness rate; While the K value is big, noise is easy to be introduced; An optimal K value can be found through the cross validation and K-nearest neighbor algorithm, which makes a low incorrectness rate of pattern recognition.
出处
《西安工业大学学报》
CAS
2015年第2期119-124,141,共7页
Journal of Xi’an Technological University
基金
国家自然科学基金青年项目(51105291)
关键词
K近邻算法
交叉验证
拒识率
误识率
K-nearest neighbor algorithm
cross-validation
refusal rate
incorrectness rate
