Research On Nuclear Power Plant Fault Diagnosis

Nowadays,fossil fuel dominate the energy structure,pollution is getting worse.At the same time,sustainable development become the focus of the research because of the characteristics of non-renewable energy.Nuclear energy is very clean and environmental,nuclear power plant has a wide range of applications in land power generation,marine and space technology.During the running of nuclear power plant,it will generate faults due to the performance degradation and improper operation.It is of great importance for effective fault diagnosis of nuclear power plant.The fault diagnosis of nuclear power plant is based on a large number of normal data and fault data.However,due to the particularity and confidentiality of nuclear power plant,it is difficult to obtain normal samples and fault samples through experiments.If we establish a mathematical model for simulation,a large number of normal samples and fault samples can be obtained.On this basis,further research on fault diagnosis algorithm can be carried out.At first,this paper establishes the model of marine nuclear power plant.We respectively establish the primary circuit model and secondary circuit model through the modular thought.Then,we use the strategy of constant secondary circuit steam pressure to design controller learn from thermal power generating unit.We get whole marine nuclear power plant model for fault diagnosis.Then,we set up faults inside the model to simulate,the three kinds of faults are deterioration of steam generator heat transfer,control rod ejection and small crevasse accident.Meanwhile,a comparison was made with off-condition conditions to explain the change of mechanism characteristics when the fault occurred,laying a foundation for fault diagnosis in the next step.In numerous nuclear power plant parameters,we select 19 characteristic parameters for fault diagnosis.Through the contrast experiment to choose the best feature extraction algorithm for dimension of 19 characteristic parameters.On the basis of dimension reduction process,we use support vector data description method for fault diagnosis.At the same time we explain the advantages and rationality of selection method through contrast experiment.Finally,we use long and short term memory network that has great performance for time series data to diagnose the data of deterioration of steam generator heat transfer.The results show that the long and short term memory network has great performance for fault data include dynamic variable condition.