A Study On Process Data Mining Technique Based On Support Vector Machines

Support vector machines and data mining are two popular research topics studied by domestic and overseas scholars nowadays. Data mining has achieved lots of accomplishment in business world and other fields. However, process data mining combined with SVM has few successful examples in industry. In this paper, some data mining algorithms combined with SVM are proposed and applied in industry. There are some main works in this paper as rendered below.First of all, a method which combines with SVM and KPCA is proposed and applied in the compound fertilizer production which is so complex that the traditional methods could not get a good result on modeling of fertilizer's component. Using support vector regression directly, data preprocess and kernel parameters selections are hard problems. In this paper, a KPCA-SVR method is proposed. The kernel principal component analysis is used to process data in order to get nonlinear principal component and get noise data off. Then an improved parameters selection method is introduced to predict the final component of the compound fertilizer. Simulation results based on practical industrial data show the effectiveness of the proposed modeling approach.Secondly, a new way of association rules extraction based on SVM is proposed in this paper. The SVC and data description is used to analyze the sample data, and the obtained support vectors are used to get the association rules in this method. It takes advantage of the abilities of SVM which can deal with limited samples and nonlinear data and have a good generalization performance. At the same time, it overcomes the unintelligible problems of SVM's classifiable function. And the program efficiency is improved by introducing the classic SMO algorithm. Simulations based on industrial data have been done and the results showed great effectiveness of this proposed modeling approach which could provide a novel idea to get the association rules.Thirdly, industrial process data mining software which is used in a WSA process and developed by myself is introduced. It includes several main functions, such asassociation rules, classification, clustering and regression. Other functions like three dimension display and online direction are also involved in the software. The data mining software makes it easy to control the process.