Future State Prediction Of Nuclear Reactor Parameters Based On Thermodynamic Best Estimation Procedure And Elman Neural Network

China is currently in the context of industrial and energy restructuring.Nuclear power,with its high energy density,good stability and low fuel transport,has played an important role in meeting national energy supply security,implementing structural reform on the energy supply side and achieving energy restructuring,with the proportion of installed nuclear power gradually increasing and the individual capacity of nuclear power units increasing.Transients occur when the state of a nuclear power plant(NPP)changes from a normal state to an abnormal state.The study of future state prediction of nuclear reactors during transients is of great importance for the safe and efficient operation of nuclear power units.Transient prediction is the prediction of the operating parameters of a nuclear power plant in order to perform appropriate operations before the transient enters a critical state.Prediction of the future state of a nuclear power plant(i.e.transient prediction)can be used to estimate the time available to control a transient and to assess the most efficient operation before the transient enters a critical state.In general,parameter prediction can be achieved by simulation model-based or data-driven approaches.For the simulation-based approach,mathematical and physical methods are generally used for modelling and simulation calculations,so that they are well adapted to different operating conditions,while simulation models such as the RELAP5 model are more time-consuming and complex to construct due to the complexity and high non-linearity of nuclear reactors.For data-driven approaches,which are independent of the simulation model,the three main methods associated with them are: unsupervised learning,semi-supervised learning(SSL)and supervised learning.Supervised learning methods are now widely used for parameter prediction in nuclear power plants.The advantages are that only operational data is required for learning and that they have a powerful ability to approximate non-linear functions.In view of the current state of international research,this paper interprets the parameter prediction method through the following technical route and chapter arrangement: proposing a novel method for future state prediction-applying the method to the studied holistic bench to obtain a prediction model-performing prediction validation of the model.The approach to future parameter prediction adopted in this paper is a combination of mathematical modelling and supervised learning methods,based on an Elman artificial neural network and the thermal hydraulic best estimation program RELAP5: the Elman neural network is used to make online single-step predictions for experimental values,and this is used as a benchmark to correct the key parameters of RELAP5,so that the prediction accuracy of the corrected RELAP5 model is improved.The prediction accuracy of the modified RELAP5 model is improved.For the validation of the model and method,the data used in this paper are from the accident simulation experiment B3.1 of the German holistic effects test rig PKL-III.The holistic effects experiment is an important part of the development of advanced reactor systems and the validation of accident analysis procedures by obtaining the overall system behaviour and response of the plant system through experiments on the holistic effects test rig.This paper presents a new method for predicting the future state of nuclear reactor parameters by combining the time-series data prediction of neural networks with the RELAP5 thermal-hydraulic best estimation procedure,which improves the computational accuracy of the original RELAP5 model and enhances its future prediction capability through the "predict-correct-parameter-reforecast" approach.prediction capability.