Research On ELM-based Fault Diagnosis Method For Thermal System Dynamic Process

Fault diagnosis of thermal system is of great significance for increasing loading security and efficiency of power units.Under the power grid-centred load dispatching mode,a large-scale coal-fired power plant often works under variable load conditions,which leads to the fault diagnosis of thermal system becoming more difficult.At present,lots of successful fault diagnosis methods for the power plant thermal system have been proposed,but for load-varying dynamic conditions,few effective diagnostic methods have been found.Therefore,fault diagnostic method based on Extreme Learning Machine(ELM)algorithm for the thermal system load-varying dynamic process is deeply studied based on abundant operation parameters,simultaneously,experimental tests are carried out with the 600 MW supercritical power plant simulator system.Accurate prediction of normal value of fault characteristic parameters is the most important link of fault diagnosis progress for the dynamic thermal system.In this paper,the prediction method of fault characteristic parameters is studied with deep understanding of ELM algorithm principle.Taking the high pressure heater of 600 MW supercritical power plant as the object investigated,the ELM prediction model is built,which can achieve accurate prediction of the normal values of fault characteristic parameters for the high pressure heaters in dynamic condition.Furthermore,in order to increase the performance of ELM prediction model,the Particle Swarm Optimization(PSO)algorithm is used to optimize the weights and thresholds of the ELM prediction model,and the PSO-ELM hybrid algorithm is built.The simulation experiments prove that the PSO-ELM model has good generalization performance and increases the prediction accuracy of ELM model.Building fault diagnosis model is the core work to achieve on-line fault diagnosis.In this paper,feature samples of the typical faults for the high pressure heater system are extracted with the 600 MW supercritical power plant simulation system,which is used to build fault fuzzy knowledge library.Then the fault diagnosis model for the high pressure heater system is built based on ELM algorithm and the off-line training and test of the fault diagnosis model is accomplished.Real-time diagnosis procedure is developed with MATLAB platform,and detailed fault diagnosis simulation tests are carried out in the power plant simulation system.Experimental results show that the fault diagnosis model proposed in this work can achieve on-line real-time diagnosis rapidly and accurately for the typical faults of thermal system under steady state and load-varying condition.