Research On Fault Intelligent Diagnosis Of Power Plant Thermal System Aided By Simulation Technology

Faults in a thermal power unit greatly influence the operation safety and economy of theunit. Therefore, it is of great significance to carry out fault diagnosis study to the power unit.Many successful works have been done in this field but most of the works focus on turbinevibration faults diagnosis based on shafting supervision. It is noticed that faults happened in thermal system of thermal power units are also in aquite high proportion, but less successful onsite application of fault diagnosis system to thethermal system has been found at home and abroad. With the concept of comprehensivemulti-parameter analysis, the fault diagnostic method to the thermal system, especially to theregenerative cycle system, is investigated. Thermal power unit is a very complex system consisting of many sub-systems in strongcoupling and is often working in various working conditions. Therefore, it is difficult to builtand consummate the fault knowledge bases of the thermal system from operation experiences,which puzzles people and holds back the technical advancement in thermal system faultdiagnosis and its application. Thanks to the progress in power simulation technology these years, high-fidelity thermalsystem dynamic model and powerful simulation support system is possible to be used in faultdiagnosis investigation of thermal system. In this paper, a new method of fault sample knowledge abstraction is put forward, whichis to abstract all fault characteristic rules of the thermal system from a high-fidelity dynamicsimulation model of the system. In order to get the fault characteristic rules, high-accuracydynamic models for thermal system main facilities, including the double-channel condenserand the surface heater, are developed and examined in depth. Based on this model, the faultcharacteristic rules of the double-channel condenser, low-pressure heater system andhigh-pressure heater system for a 300MW thermal power unit are investigated. The faultknowledge bases, which are more complete, more reliable and closer to engineeringapplication, are achieved from this way. The theory and method of thermal system fault intelligent diagnosis is another importantcontent to study in the paper. The main contents involved in fault intelligent diagnosis,including symptom fuzzy expression, artificial neural network (ANN) fault diagnosis methodand fault fuzzy pattern recognition method, et al. are discussed in detail. Two types of fault symptoms, trend symptom and semantic symptom, are adopted forthermal system fault diagnosis. Their fuzzy calculating methods are explained and theirintegrated calculations under different cases are given for the first time. An improvedmembership function is used for fault diagnosis with fuzzy pattern recognition method, whicheffectively overcomes the demerits of the existing function forms and improves its versatilityand fault recognition effect. The back-propagation (BP) neural network based fault intelligent diagnosis method isemphatically studied and a new-style radial basis function (RBF) neural network faultdiagnosis method is also attempted. A new neural network training method for BP networkcalled "learning rate self-adaptive adjustment based on constant error correcting rate" is putforward which greatly shortens the network convergence time and is favorable for real-timeon-line fault diagnosis. In order to realize slight and incipient fault diagnosis correctly and timely for thermalprocess, an intelligent fault diagnosis method is put forward for the first time by jointapplication of symptom zoom technology (SZT) and fuzzy symptom representation. Thecomplexity of the thermal system fault knowledge base can be effectively reduced with thismethod and the slight and incipient fault can be correctly diagnosed timely and accurately. To verify the maturity and allsidedness of the thermal system fault knowledge bases, andthe validity of the fault intelligent methods studied in this paper, an...