Research On Locating And Predicting Leakage Of Four Tubes In Power Station Boiler Based On Deep Learning

With the increase of thermal power units’ deep peaking operation time and the policy background of "carbon peaking and carbon neutral",the safe and stable operation of thermal power units becomes more and more important.In the actual operation of power plant boilers,"four tube leakage" is not only a major accident that causes sudden shutdown of power plants,but also occupies a large proportion of equipment failure in thermal power plants.Therefore,whether from the frequency of occurrence,or the severity of harm,its prediction and location is a matter of great concern to power plant personnel.In this paper,we study and analyze the problem from two perspectives of localization and prediction,and use a combination of acoustic detection and deep learning to locate the leakage acoustic signal and improve the accuracy of leakage point location.On the basis of the localization,it is hoped that the deep learning neural network method is used for the leakage prediction of the power plant boiler operation data.For the acoustic leak localization part,the relevant experimental bench is built,and compressed air is used as the leak medium to obtain the time delay value after the microphone array acquisition by PC calculation time delay estimation;the acoustic feature extraction(MFCC,CEEMADAN)and recognition(SVM,CNN)methods of the leakage signal were compared and validated to obtain the recognition accuracy of the four combined methods.The principle and implementation procedure algorithm of geometric localization method are given;a localization algorithm based on LSTM deep learning neural network is given,and its implementation process is described.Under the experimental conditions,the sampling rate increases and the localization accuracy gradually improves;with the increase of the sampling distance,the average accuracy of the localization distance does not change much,and the fluctuation of the localization distance data increases significantly,i.e.,the distance data deviates from the average.The accuracy of predicting location using different neural networks,with the LSTM algorithm being slightly more accurate than the other two(BP,CNN);the localization accuracy of the neural network increases with the amount of data.On the basis of leakage location,the data of coal-fired boiler are preprocessed and reduced in dimension for the four-tube leakage prediction part of power plant boiler;explain the leakage prediction process of four tubes of power plant boiler;the SSA-based optimization algorithm is used to predict the parameters characteristics of power plant boiler load,import and export temperature of each heating surface,flow rate,pressure,etc.,and the error of each parameter prediction model is within 10%;by calculating the deviation of series parameters characteristics,the four-tube leakage health index of boiler in thermal power unit is finally obtained,and the possibility of leakage of boiler is predicted by calculating the deviation of series of parameter characteristics.