Research Of Diagnosis And Prediction For Turbine Thermal System Fault By Improved BP Algorithm

In this paper, the application of improved BP algorithms, which include improved conjugate grads BP algorithm, adding momentum BP algorithm, self adapting BP algorithm, recurrent compose BP algorithm and subsection optimize BP algorithm, is discussed for turbine thermal system, and the improved conjugate grads BP algorithm and recurrent compose BP algorithm are chose to develop the software of fault diagnosis and fault prediction of turbine thermal system for ChongQing Power No.21 Unit. During the process of software development, the new samples of fault diagnosis and fault prediction are collected, the knowledge database of fault diagnosis and fault prediction is extended, Knowledge acquirement module, diagnosis prediction module and interpretation module are established, the software of fault diagnosis and fault prediction is programmed by Visual C++, and simulation tests of frequent faults diagnosis and important character parameters prediction for these subsystems of high-pressure ,feed water deaerator and condenser, are carried out. The fault diagnosis simulation tests show that the training velocity of adding momentum BP algorithm, self adapting BP algorithm and improved conjugate grads BP algorithm are faster than traditional BP algorithm, among others improved conjugate grads BP is best because it can better solve the problem which fall in local least point during faults diagnosis of the complex turbine thermal system, further more the fault diagnosis precision is more accurate, can be judged whether the faults happen and what the fault severity degree is. The tests of prediction simulation indicate that the recurrent compose BP algorithm and subsection optimize BP algorithm all can predict the fault character parameter of turbine thermal system. The prediction precision with the recurrent compose BP algorithm is exact, because it adds the link weight between the different nodes of the same layer and between the nodes of the input layer and the output layer, which enhances the conjunction ability of inner nodes, and it adopts linear prompting function, which overcomes the prediction saturation using traditional BP algorithm. The net training velocity and precision can be highly improved by using subsection optimize BP algorithm, but it has need to avoid "over imitation". The research can provide reference for realization of the fault diagnosis and fault prediction for turbine thermal system in order to ensure the safe operation and economical operation of power plant.