调速阀故障诊断的AR双谱定阶方法比较

Comparison of Different Methods of Determining the Order of Autoregressive Bi-Spectrum

为解决故障诊断中确定最优的自回归(AR)模型阶数的问题,运用最终预测误差(FPE)、阿凯克信息准则(AIC)、贝叶斯信息准则(BIC),以及奇异值分解(SVD)的切片法和Frobenius法共5种定阶方法对调速阀的故障进行自回归模型定阶实验.结果表明,FPE,AIC,BIC及SVD切片法确定的阶数较低,而用SVD Frobenius法确定的阶数较高.通过不同阶数、不同故障的调速阀故障诊断实验可知,用SVD Frobenius法建立的AR模型效果优于其他方法.

In order to determine the optimal order of autoregressive （AR） models, the 5 methods including final prediction error （FPE）, Akaike＇s information criterion （AIC）, Bayesian information criterion （BIC）, singular value decomposition （SVD） based Slicing, and Frobenius are used in the corresponding experiments for the faults of speed-regulating value. The results have revealed that the selected orders of the AR model using the FPE, AIC, BIC, SVD slicing methods are lower than that obtained using the SVD Frobenius method. It has been proven that the AR model obtained by the SVD Frobenius method is more efficient than those obtained using other methods under the different fault diagnosis experiments for speed-regulating value.