A Feature Selection Method Based on Variable Weight in Fault Isolation

In the fault isolation of multivariable features, it is very important to select effective features from hundreds of features for feature classification. Fault isolation based on intelligent model has received considerable attention in academic research, most of which optimize the classification model, but little research worked on feature selection. This paper focuses on the study of the feature selection method to reduce the selected feature. The generic information entropy is usually applied to measure the dispersion of each feature and to construct the Entropy-weight of the feature. To emphasize the discriminability of feature and modify the entropy weight, this paper proposes a method, constructing the feature weight by partial F value based on multivariate hypothesis testing to measure the effect of each feature on the difference between different feature sets. Then the modified weight is utilized as criteria for selecting the most effective feature for fault isolation, reducing the number of selected features to achieve same effect of fault isolation and relieving the computational pressure of the classifier. The performance of hydraulic equipment fault isolation illustrates the effectiveness of the presented method.