Study On Machine Learning Fault Diagnosis Of Power Station Fan

Fan is one of the important auxiliary equipment of thermal power generation unit.Its safe and reliable operation is directly related to the stability,economy and safety of the unit.With the rapid development of artificial intelligence technology,the research on intelligent identification,diagnosis technology and fault trend prediction of power production equipment fault has been gradually carried out.Based on data statistics and machine learning technology,this paper studies intelligent fault diagnosis,identification and fault early warning of fans in power plants,so as to provide support for reducing losses caused by faults,avoiding fault deterioration and improving fault treatment efficiency.The main work of the thesis includes:Research in machine learning are analyzed as the main line,the method of fault diagnosis of several of the fault diagnosis method based on machine learning are summarized and compared the characteristics of each method,this paper used clustering and least squares support vector regression for acquisition of fault diagnosis and trend prediction data,and analyses the power station of the type,structure and fan several kind of typical fault.Fan vibration signal based on plant data,the use of wavelet packet technique to feature selection of data,the index as a fan of the characteristic of time domain,the calculation of wavelet packet energy as the characteristic,application of DET evaluation technology,Pearson correlation analysis,improved feature extraction conditions of hybrid technology to extract features,application of K Means clustering will fault classification,the simulation show that this method has lower error diagnostic rate and higher diagnostic accuracy.The trend prediction method of fan crack fault is studied,the fault characteristic quantity represented by induced draft fan is taken as the input,and the average characteristic quantity represented by trend characterization index which reflects the characteristics of original data is taken as the output.Aiming at the problem of large residual error between the predicted value and the actual value of the least squares support vector regression(LSSVM),a particle swarm algorithm-least squares vector regression(PSO-LSSVM)parameter optimization scheme was proposed.