Clustering Method Of Time Series Symbolization On Mine Hoist Fault

With the rapid development of modern industrial science and technology, andmore complexity of the structure of equipment system, the safety and reliability ofthe system has been focused on. Many researchers have involved in the research oftechnology on fault diagnosis. The hot spot in the current study covers how to realizethe lower complexity and better online diagnosis ability.The relevant technology of the time series is studied extensively at present,some of which in fault diagnosis have been applied successfully. This thesis wascarried out from the metric of similarity distance of time series and clusteringanalysis in fault diagnosis technology. The concrete content is as follows:Firstly, this thesis studies the common algorithm of the metric of similaritydistance on time series. In terms of the lower matching efficiency of traditionalEuclidean distance and the higher time complexity of DTW distance metric, thealgorithm of similarity metric of time series symbolization is proposed based on thelongest common subsequence without support of timeline scaling and translationproblems in the comparison of similarity. The flexible matching parameters isinputted into the algorithm, which makes the sequence can reduce the noiseinterference. In addition, the matching path is specified, thus, the time of thealgorithm is reduced.Secondly, on the basis of the algorithm of similarity distance metric, aalgorithm of time series clustering is proposed, realizing the clustering analysismodel based on the distance of the characters. Algorithm using E-LCS distance as agist of the clustering algorithm in crude similarity division establishes a membershipmatrix and realizes clustering analysis of time series, combined with the fuzzyclustering thought. A lot of public data sets of experiments show that under thecondition of the limited increment of time-consuming of the algorithm; the accuracyof the clustering is improved greatly.Finally, the clustering algorithm is applied to improve the system of faultdiagnosis in this thesis, combined with the comprehensive evaluation commonlyused in the algorithm of fault diagnosis at present. Through the experiment, thewhole model is validated and evaluated, which proves that the algorithm ofELCS-Cluster could be effectively improved the accuracy of the hoist faultdiagnosis.