关键词: 特征选择, ReliefF算法, 拉普拉斯分数, 改进灰狼优化, 故障诊断

Abstract: To select the most discriminating feature from the original high-dimensional feature space and improve the bearing fault diagnosis accuracy,a fault feature selection algorithm combining Filter and improved Gray Wolf optimization was proposed.For the original vibration signals of rolling bearings,a time-frequency domain feature extraction strategy based on Hilbert-Huang transform was used to establish a high-dimensional sensitive feature set.Then,a hybrid Filter method consisting of ReliefF and Laplacian Score (LS) was used to evaluate the relevance of the original feature set and quickly select important features.Thus,a pre-selection of feature set was completed.The Improved Grey Wolf Optimization (IGWO) was introduced to perform secondary selecting on the pre-selected feature set.At the same time,the parameters of support vector machine model were optimized.The experimental results showed that the proposed method significantly improved the diagnostic accuracy of the classifier model and effectively reduced the feature dimension of the fault data set.Compared with similar methods,the proposed method had better comprehensive performance.

Key words: feature selection, ReliefF algorithm, Laplacian score, improved grey wolf optimizer, fault diagnosis