Reserch On The State Of Early Warning And Health Based On The Key Equipment Of Nuclear Power Plant

At present,various in-service equipment of nuclear power plants has different degrees of aging.In order to prevent various unpreventable failures of the equipment,to ensure the safe and stable operation of nuclear power plants.This paper conducts related research on the state early warning and health of key equipment in the primary loop of nuclear power plants,which can ensure the safe and stable development of nuclear power,while reducing costs,and has important application value for the economic and reliable operation of nuclear power plants and the future development of intelligence.In this paper,data mining,neural networks and other machine learning theories are used as the main technical means to conduct in-depth research on the state warning and health prediction of key equipment in nuclear power plants.Entropy weight method and grey correlation degree are used to quantify the degree of state deviation of key equipment,combined with the improved Gaussian mixture model to determine its early warning state,and realize state early warning.Then evaluate the health of key equipment,and use the health status obtained from the evaluation to predict the health of the equipment.The main contents are as follows:First,an improved GMM-GRA-EWM early warning method based on multi-characteristic parameter synthesis is proposed.Based on the mechanism model of the average coolant temperature and the automatic adjustment principle of the control system,the mechanism of the dynamic operating characteristics of the small leakage of the primary circuit is analyzed,and the early warning characteristic parameters are determined.According to the determined early warning characteristic parameters,a multi-parameter comprehensive early warning model is established by the method of combining entropy weight method and gray correlation degree.Correlation analysis and improved Gaussian mixture model algorithm are used to effectively learn the statistical characteristics of a large amount of data,so that the early warning threshold has an adaptive ability under different working conditions.Then,a health model based on similarity measures was established.Because the Mahalanobis distance can make an excellent description of the similarity of multivariable time series data,the model uses the Mahalanobis distance to describe the health status of the equipment,and on the basis of this Mahalanobis distance,the entropy weight method and the analytic hierarchy process are used to analyze the It is improved to further improve the accuracy of the health model.Secondly,a health prediction model based on convolutional neural network(CNN)and long short-term memory neural network(LSTM)is designed.The model uses the efficient data feature extraction ability of convolutional neural network to determine the related data of health feature parameters.Perform multi-layer convolution calculation and pooling stacking to extract data features related to health.Since the long and short-term memory neural network has good processing capabilities for the long-term dependence of time series,the data features extracted by CNN are used as the input information of LSTM to obtain the health prediction model of CNN-LSTM.Finally,through the typical failures of the nuclear power plant simulation system,the state warning technology and health prediction technology studied above were verified by examples.The verification results show that the state early warning technology proposed in this paper can provide timely and effective early warning of the leakage of key equipment in the primary loop;at the same time,the health and prediction model constructed in this article can better describe the health status of the equipment and accurately predict the primary loop The health of the system equipment changes trend,and the prediction error is small.It proves the feasibility,effectiveness and practicability of the state warning technology and health prediction technology studied in this paper.