Research On Fault Diagnosis Method For RCS Of Nuclear Power Plants Based On Manifold Learning

There are potential radioactive hazards in the operation of nuclear power plants,and the consequences of radioactive leakage will be extremely serious.The particularity determines the safety requirements of the nuclear power plant system and equipment.Therefore,it is of great significance to improve the safety and reliability of nuclear power plants by studying the technologies of early fault detection and fault diagnosis during anomalies occur in a nuclear power plant,it is very important to determine the system operation status as soon as possible and give the correct judgment to the operators.At the same time,the operating status of the system can be determined as soon as possible and the operator can be judged correctly.In this paper,a set of intelligent fault diagnosis system is developed for the reactor coolant system of nuclear power plant,which solves a series of problems such as early fault detection and diagnosis of nonlinear system.The diagnostic system is divided into four sub-modules according to functional logic,which are as follows: fault feature dimension reduction and clustering,early fault detection,pattern recognition of fault types,and fault degree approximate evaluation module.By analyzing and comparing the effects of different methods,the best datadriven method is selected for each submodule according to functional requirements and finally integrated under the same data platform for unified scheduling.The main work of this paper is as follows:(1)Through artificial dataset and Nuclear power plant operating data test,the data dimensionality reduction,feature extraction and clustering ability of nonlinear manifold learning method and linear manifold learning method are comparatively analyzed.The LTSA manifold learning algorithm with the best overall performance is determined as the feature clustering method.(2)The non-linear LTSA method replaces the traditional linear PCA method,and is applied to the fault detection of reactor coolant system of NPP.A fault detection method of reactor coolant system based on improved statistics is proposed,which can detect the failure early in the fault and reduce the false alarm rate.(3)The feature clustering ability of manifold learning method was studied and combined with SVM and K-nearest neighbor classifier respectively to establish two fault diagnosis models and realize the recognition of fault types.A clustering-reclassified LTSA-KNN fault diagnosis method based on manifold learning method and K-nearest neighbor classifier is finally determined through experiments.(4)A fault degree evaluation method based on BP neural network is proposed to approximate the fault degree of partial fault of the reactor coolant system.Through the Visual Studio 2010 platform and Matlab mixed programming man-machine interface-friendly manifold learning fault diagnosis system.The PCTRAN simulation software was used to simulate some of the reactor coolant system faults,which verified the feasibility and effectiveness of the proposed method.