Research On Fault Diagnosis Of Supercharged Boiler Based On Deep Learning

The marine pressurized boiler is the power source of conventional steam power system,which is the key equipment to ensure the safe and reliable operation of power plant.If the fault of the supercharged boiler is not detected and eliminated in time,it may cause the paralysis and failure of the entire steam power system,and even cause great harm to the entire ship.Therefore,it is of great practical significance to study the fault diagnosis of marine supercharged boiler.In this paper,the simulation method and deep learning method are combined,and the simulation model of supercharged boiler is established by using GSE simulation software,and the occurrence of overheating fault of superheater caused by seven kinds of induction reasons and jam fault of auxiliary steam turbine during load reduction process are simulated.The fault sample data were collected,and the fault sample database of superheater overheating was established.Based on the long-term and short-term memory cycle neural network threshold method,the fault detection and fault early warning are studied for the overheating problem of super-heater of supercharged boiler.The same method is used to detect the jam fault of auxiliary steam turbine inlet valve in the process of load reduction.Deep fully connected neural network model,one-dimensional convolution neural network model and two-dimensional convolution neural network model are used to study the fault location of superheater over temperature.The common model evaluation indexes in machine learning are introduced to compare the performance of three fault diagnosis models.The results show that the threshold method based on long-term and short-term memory circulation neural network can accurately detect the overheating fault of superheater,and can issue early warning information in advance.The precision rate of fault diagnosis model is 98.39%,the recall rate is 97.56%,and the F1 score is 97.97%.For the fault detection of auxiliary steam turbine inlet valve jam in the process of load reductio,the neural network training is difficult,and the accuracy of the diagnosis model is reduced,the precision rate is 94.36%,the recall rate is 93.61%,and the F1 score is 93.98 %.For the fault location of superheater overheating problem,the two-dimensional convolution neural network model has the highest reliability of diagnosis results,the recall rate,precision rate and F1 score are more than 95%,which are higher than the other two neural network models,proving that the method has high reliability.