Feature Selection Method Based On Cuckoo Search And Its Application In Bearing Fault Diagnosis

In recent years,data in various fields around the world have shown a surge.Massive and high-dimensional data appear in all aspects of people’s lives,bringing convenience to life and new challenges to researchers.Feature selection processes the original data by removing irrelevant and redundant features,which achieves the purpose of improving the accuracy of classification and reducing the cost of calculation.In this paper,the feature selection method based on cuckoo search is studied and improved,and the improved feature selection method is applied to bearing fault diagnosis.The main work of this paper is divided into the following three parts:(1)In order to solve the initialization random problem of feature selection algorithm based on cuckoo search and optimize the classification performance,a feature selection method based on mutual information enhanced cuckoo search is proposed.This method first proposes a feature weighted initialization method based on mutual information in the cuckoo population initialization stage to improve the quality of the initial population.In addition,the local spiral development strategy is presented in the search stage of the cuckoo search algorithm,and the spiral search strategy is introduced in combination with the moth fire algorithm to enhance the search ability in the later stage.Experiments on UCI data sets show that the proposed two improved strategies can effectively reduce the number of selected features and classification error rate of feature selection algorithm based on cuckoo search.(2)A feature selection algorithm based on clustering hybrid binary cuckoo search(CHBCS)is proposed.This method combines the clustering information between features to further improve the initialization method and optimize the convergence speed of the algorithm.Firstly,a clustering hybrid initialization method is proposed to effectively delete redundant features by using the clustering information between features.Secondly,a mutation strategy based on Levy flight is presented to effectively use high-quality information to guide subsequent search.In addition,a dynamic discovery probability is proposed based on population sorting,which can effectively retain high-quality solutions and accelerate algorithm convergence.The experimental results on UCI datasets show that the clustering hybrid initialization method can effectively improve the classification performance of the algorithm.The CHBCS algorithm can effectively reduce the number of features and improve the convergence speed of the algorithm under the premise of improving the classification accuracy.(3)In order to solve the problem of low accuracy caused by irrelevant and redundant features in rolling bearing fault diagnosis,a CHBCS-KNN fault diagnosis model is constructed and tested on the CWRU bearing dataset.Firstly,Hilbert-Huang transform is used to extract fault features in the signal processing stage.Secondly,the CHBCS feature selection method is used for feature selection.Finally,KNN classifier is used for fault diagnosis.The CHBCS algorithm is compared with other nine optimization algorithms.The results show that the CHBCS algorithm can effectively identify a variety of bearing fault types,and is superior to other optimization algorithms in terms of diagnostic accuracy,number of features,convergence speed,and stability.Therefore,the proposed CHBCS-KNN fault diagnosis model has good application value in solving the problem of rolling bearing fault diagnosis.