Neural Network Fault Diagnosis Approach For Power Plant Thermal System Under Load-Varying Dynamic Transition Process

Fault diagnosis for the power plant thermal system is of great significance to improve the safety and economy of the whole power unit. A thermal power unit often works in dynamic transient condition in order to meet the changing need of the power load. At present, most researches on power plant fault diagnosis are for steady state load conditions. The existed fault diagnosis methods are not well adapted to the dynamic transition process. Coupled with the complex structure of the thermodynamic system and faults diversity, the application of fault diagnosis system has been greatly limited.This paper put forward a transient fault diagnosis method for power plant thermal system based on two-stage neural networks. Among them, the first neural network adopted an Elman neural network with time-delay inputs to predict the expected normal values for the fault-related feature variables, and the second neural network used a BP network to identify the fault type. Fault symptom optimization technique was also used to improve the fault diagnostic effect for faults of varying severity under changing load conditions.By using a full-scope simulator of a600MW supercritical power unit and taking its feedwater heater system as the object investigated, the prediction model for the expected values of the fault feature variables was built, trained and validated by large amount of historical operating data, including typical steady-state load points and load-varying dynamic transition processes. By simulation study of typical faults, the fault fuzzy knowledge set for the feedwater heater system was built, and the BP neural network fault diagnosis model was then trained off line. The fault diagnosis program was developed with MATLAB software. By communicating with the simulator, detailed fault diagnosis simulation tests were carried out. It was shown by tests that the suggested method can achieve good diagnosis results for the power plant thermal system both at steady state conditions and load-varying dynamic transition process.