Fault Diagnosis Of Combined Cycle Unit Heat Recovery Boiler Based On SVM And CNN

Hangzhou Banshan Power Plant(pilot power plant)of Huadian Group takes the opportunity of building a digital power plant to put forward the demand of building a fault diagnosis system for the heat recovery boiler of No.7 combined cycle unit.The research content of this paper starts from the actual demand,and the research object is the heat recovery boiler system of gas steam combined cycle unit.Because of the multi-mode and time-varying in the power production process of combined cycle units,it is of great significance to establish the detection of normal and abnormal working conditions by using CNN algorithm and to apply SVM model to fault diagnosis for improving the accuracy of fault diagnosis and improving the safety and reliability of combined cycle units.The main research contents of this paper are as follows:Firstly,the actual operation data of the heat recovery boiler system of the No.7combined cycle unit in the pilot power plant are collected,"cleaned" and standardized.Combined with the characteristics of the actual sample data,the model based on SVM is extended and applied.A heat recovery boiler fault diagnosis system based on multi-classification SVM is developed.In order to reduce the false positive rate of the system,a binary SVM model group decision function based on "confidence" is proposed.The performance of the fault diagnosis system is verified in the test set of actual sample data.Secondly,in order to make up for the deficiency of fault diagnosis ability of the heat recovery boiler fault diagnosis system based on multi-classification SVM for unknown fault diagnosis,the heat recovery boiler fault detection system based on CNN is developed.A method to transform vector data into matrix data--"Industrial Data Camera" is proposed.It is possible to input the operation data of heat recovery boiler to CNN.The training set of actual sample data is used to train CNN,and the performance of its detection system is verified on the test set.Finally,combined with the characteristics of the above fault diagnosis system and fault detection system,a heat recovery heat boiler fault diagnosis system combining SVM and CNN is developed,and a binary SVM model group automatic updating algorithm is proposed with manual marking mechanism when the unknown fault is detected.The heat recovery boiler fault diagnosis system as a sub-module,embedded in the thermal power generation equipment running status remote monitoring platform to run,the pilot power plant heat recovery boiler system fault diagnosis practical application,and test to verify its performance.