The Research Of Clustering Algorithms And Its Application On The Fault Diagnosis Of Wastewater Treatment Technics

Clustering is an important mining method and mining task of data mining. It finds the comparability among the data from the database, and classifies the data to make different data in different kind as much as possible, and the data in the same class are as similar as possible, i.e." birds of a feather flock together ", thus optimize the implicit useful information or knowledge in the inquiry of the extensive database and discovery data, there is extensive application in a lot of fields in the data clustering.Wasterwater treatment process is a biological process that contains many complicated process technics, it is hard to run safely and stably for long time. The main intelligent technology used for fault diagnosis in wasterwater treatment's daily running process in our country is knowledge-based expert system. One of the difficulties inherent to the development of a knowledge-based system is to obtain the knowledge base. The establishment of the knowledge base is mainly due to the few operators'or experts'experience, so it causes knowledge acquisition bottleneck problem. The wastewater treatment would generate a large amount of data after running for some times. Clustering analysis is one kind of non-surveillance classification technology which could good at analyzing the data which have few prior knowledge, therefore we proposed using clustering analysis technology to analyze the wasterwater treatment's history data, through analyzing the clustering results, to generate a set of inference rules for supplementing the knowledge base of fault diagnosis system of wasterwater treatment technics.After researching and analyzing the existing clustering algorithms, we presented two clustering algorithms based on genetic algorithm and based on the nearest neighbor clustering respectively. Next, we used the clustering algorithm to the wastewater treatment plant's history data, and do some preliminary discussion to the establishment of fault diagnosis rule of fault diagnosis systemThe main contributions of this dissertation are summarized as fellow:â‘ We introduce some elementary knowledge of clustering technology and genetic algorithm as the basis of this dissertation. The coding mode and operation in the genetic algorithm are discussed and the choice of control parameters is analyzed also.â‘¡This paper is engaged in the hybrid algorithm of K-means algorithm and genetic algorithm. We presented an improved cluster algorithm based on genetic algorithm. It can enhance convergence speed and solve clustering problems.â‘¢This paper proposes a novel clustering algorithm which is based on the nearest neighbor clustering and genetic algorithm. The algorithm includes two stages, begins by running the nearest neighbor clustering algorithm to establish the set of original clusters using the nearest neighbor method by grouping very similar instances into a cluster based on some similarity or dissimilarity metrics, and then employs genetic algorithm to combine original clusters and obtain the near optimal result. In this algorithm , the number of clusters is not known a priori. We take some artificial synthesis data sets to test this algorithm and the clustering result is analyzed also. A two-stage clustering algorithm based on the nearest neighbor clustering is presented in this paper on basis of the prior research which could be used for partitioning clustering or hierarchical clustering.â‘£We used a novel abnormal detection algorithm based on the nearest neighbor clustering and genetic algorithm to analyze the wastewater treatment history dataset, then used outlier measure of selection the top-n abnormal item from dataset based on distance sum. After that analyzed the abnormal sample with expert's interpretation and discussed the establishment of fault diagnosis rule of fault diagnosis system.