Research On Fault Warning Of Power Station Equipment Based On Data Mining

Power station equipment as a large mechanical power equipment,its safety and reliability are very important.The traditional way of fault detection is to diagnose the equipment fault,which is a way of post maintenance.Once the power station equipment is shut down due to fault,it will bring property losses,and even casualties.Therefore,how to use other methods to achieve fault warning before shutdown is particularly important.This paper takes the induced draft fan as an example.After processing the historical data of normal operation of induced draft fan obtained from SIS system,the wavelet de-noising method is used to smooth the data.The nonparametric fault early warning model under normal operation condition is constructed by using data-driven method multi state estimation technology(MSET),and the process memory matrix is constructed by combining principal component analysis(PCA)to simplify the model.The simplified model is used to estimate the output variables of the input variables,and the residual is defined as the difference between the input variables and the output variables.The sliding window residual statistical method is used to determine the fault threshold,which is used as the judgment standard.When the residual exceeds the threshold,the fault warning information is given to remind the staff to check whether a fault has occurred.Realize the function of fault early warning.In order to verify,the data segment of induced draft fan with fault information is selected for verification.The results show that the combination of multivariate state estimation and PCA can detect the fault development process in time,and can be used for fault early warning.