Study On Condition Based Maintenance Of Equipment In Power Plant Basing On Data Mining

In order to development the efficient, energy saving, environmental protection thermalpower generator, this requires equipment maintenance regular overhaul from the traditionalgradual transition to condition-based maintenance. For the power plant, more and more datais stored in real-time database by the DAS and DCS system. With the accumulated historicaldata, it is possible to infer the power plant in the future development, trend for the futureoperation, maintenance and incident handling to provide basis for decision making.Applying data mining methods to extract the appropriate fault diagnosis knowledge fromthe real-time database is an effective way, also is a practical significance and research valueproblem.Support vector machine (SVM) is a new kind of machine learning method based onstatistical learning theory, which has many advantages. SVM solves small-sample problemsby using structural risk minimization (SRM). In this paper I using SVM based method forequipment fault detection in a thermal power plant, to analysis from feature selection andparameter optimization. I take the fault diagnosis for steam turbine generator for example, inorder to raise the efficiency of fault diagnosis of steam turbine units and consider its costsand complexity, using the correlation analysis as data pre-processor. We calculate thecorrelation coefficients between attributes, and combine with max-min distance, featureselection of failure data, then keep only one of the attributes which most highly correlates. Itis simple and classification results significantly. Then construct support vector machineclassifier, applying particle swarm optimization (PSO) to find optimal parameter.Experimental results show that SVM outperforms linear discriminant analysis (LDA) andback-propagation neural networks (BPNN) in classification performance and can be wellapplied in fault diagnosis.The quality of equipment maintenance determines the power plant can safe and reliableto run long, will directly affect the efficiency. This paper further studies the equipmentmaintenance subsystem. Through analysis the results of fault diagnosis, to measure the needfor preventive maintenance from the overall operating rate and routine maintenance costs ofequipment.