Research On The Operating Trend Prediction Technology Of The Nuclear Power Plants

Design and application of an operating trend prediction system could timely and exactly reflect the operating trend of the nuclear power plant(NPP),providing pre-alarm information to the operator.Therefore,it's of great significance to improve the safety of the NPP and ease the pressure of the operators.The paper draws a deep analysis of the operating characteristics of the NPP and the variation of the main characteristic parameters.And through the analysis of current research status in NPP operating trend prediction,it can be concluded that it's feasible to use grey prediction method or BP neural network prediction method for NPP operating trend prediction.On the basis of understanding grey prediction method and BP neural network prediction method,GM(1,1)prediction model and BP neural network prediction model are studied in depth.Aiming at the shortcoming of GM(1,1)prediction model,an improved gray GM(1,1)model was proposed through introducing block matrix calculation.The improved gray GM(1,1)model is capable of modeling for multiple original series.In addition,this study uses the technology of sample rolling updates and the technology of neural network online training to achieve a dynamic prediction model,which is called Online BP neural network prediction model.Then compare the advantages and disadvantages and determine the scope of these prediction models in time-series prediction for the main characteristic parameters of NPP with reason analysis.And research on the application of soft sensor in the key parameters of NPP based on BP neural network prediction model.For the result of the soft sensor,the reason is analyzed and the effect is evaluated.C#4.0 programming language is used to complete the design and development of the NPP operating trend prediction system in the Visual Studio2010 development environment.Some application research of the NPP operating trend prediction system,which has been built,would be given by using the full range simulation software of the NPP.The results show that when predicting multiple time-series of parameters,advanced GM(1,1)prediction model should be adopted,which is suitable for predicting the data which has a linear trend or two-order polynomial trend;When predicting single time-series of parameter,Online BP neural network prediction model should be adopted,which is suitable for predicting the data which has a linear trend,two-order polynomial trend or cycle-fluctuation trend;In addition,the system can achieve a soft sensor of the water level in SG by using BP neural network prediction model on condition of reactor load drop.