Data mining based on improved neural network and its application in fault diagnosis of steam turbine

Abstract

Steam turbine is an important equipment in the industry, especially in the electric power industry. Because of the complexity of steam turbine and particularity of its running environment, the fault rate of steam turbine is high and its harm is serious. So the fault diagnosis of steam turbine is a difficult problem. A novel approach for fault diagnosis of steam turbine based on improved neural network is brought forward, aimed at overcoming shortages of some current knowledge attaining methods. An application of artificial neural networks methodology was investigated using experimental data. Multiplayer backpropagation neural network with two hidden layers, hyperbolic tangent as the activation function and target function were studied. Neuro-fuzzy systems were also applied. Based on the ontology of neural network, the data mining algorithm for classified fault diagnosis rules about steam turbine is brought forward; its realization process is as follows: (1) computing the measurement matrix of effect; (2) extracting rules; (3) computing the importance of rules; (4) shearing the rules by genetic algorithm. An experimental system for data mining and fault diagnosis of steam turbine based on neural network is implemented. Its diagnosis precision is 84%. And experiments do prove that it is feasible to use the method to develop a system for fault diagnosis of steam turbine, which is valuable for further study in more depth.