摘要
传统滑动窗口主元分析算法在处理工程参数缓慢劣化问题时存在过适应现象,且容易产生误报、漏报的问题。为此,本文以滑动窗口核主元分析(KPCA)算法为基础提出一种基于数据块的改进滑动窗口核主元分析算法。该算法采用MATLAB软件进行数据标准化预处理,利用迭代法剔除采样数据异常值;在检测过程中以数据块为单位进行模型更新,通过调节数据块的大小,使故障诊断更加快速、准确。将该算法和传统核主元算法分别应用于风力发电机组实验数据分析,验证了本文所提出的改进算法能够更好地解决参数缓慢劣化问题。
The ＂over fitting＂ and ＂false and leakage alarm＂ problems easily occur during the application of conventional moving window principal component analysis （PCA） in dealing with slow deterioration of engineering parameters. To solve this problem, this paper proposes an improved moving window kernel principal component analysis （KPCA） algorithm based on data block, according to the sliding window KPCA algorithm. This improved algorithm carries out data standardization by Matlab and applies the iterative method to eliminate the abnormal data of the samples, then it uses the data block as the unit to update the model during the detection. Through regulating the size of the data block, the fault diagnosis will be more rapid and accurate. Moreover, this improved algorithm and the conventional KPCA method were applied to analyze the test data for wind turbine unit, the result verifies the above improved algorithm can solve the problem of slow deterioration ofnarameters in a better wav.
出处
《热力发电》
CAS
北大核心
2018年第1期100-105,共6页
Thermal Power Generation
基金
河北省自然科学基金资助项目(F2014502059)~~
关键词
核主元分析
滑动窗口
数据块
模型更新
故障诊断
过适应
迭代法
kemel principal component analysis, moving window, data block, model updating, thult diagnosis,over adaptation, iteration method
