Research On Soft Measurement Method Of Thermal Hydraulic Parameters Of Pressurized Water Reactor Coolant System Based On Deep Learning

Safety is a prerequisite for the long-term stable operation of nuclear power plants.In the field of operational safety research,finding fast and stable methods for measuring key thermal-hydraulic parameters of nuclear power plants that are less susceptible to interference is one of the most important research directions.The deep learning measurement method has become a hot research topic in recent years because it does not require additional measurement equipment and accurate mathematical models,and it has the advantages of fast response and independent measuring.In this paper,the authors propose a deep learning method for soft measurement of some key thermalhydraulic parameters of reactor coolant systems(RCS)in nuclear power plants,which provides an idea for using artificial intelligence methods to solve measurement problems in nuclear power plants.The main research contents of this paper are as follows.(1)A mechanistic model and a deep learning model of the main thermal-hydraulic parameters are established.Through the operating experience and nuclear power plant operation manual,the main research objects of this paper are identified: the inlet and outlet temperatures of the core,the pressure and water level of the steam generator.Mechanistic models and deep learning models for these parameters are developed in this paper,respectively,and a comparative study of soft measurement results is carried out using the actual operating parameters.Based on the deep learning principle and the operating trends of different parameters as well as their complexity,different neural network models are selected for different operating parameters for soft measurement studies.The measurement results were evaluated using mean absolute error,maximum error and minimum error.The feasibility and advantages and disadvantages of deep learning models for soft measurements of reactor coolant system parameters in nuclear power plants are demonstrated.(2)The robustness of the deep learning model is discussed.Considering the possible working conditions of nuclear power plants with missing parameters due to sensor failures,the authors design simulation experiments under the working conditions of missing parameters by using the robustness analysis methods commonly used in the field of deep learning,and analyzes the soft measurement results under different types of missing parameters and different numbers of missing parameters.The upper limit of parameter missing that the deep learning model can withstand is discussed and found.The robustness analysis proves that the deep learning model has good robustness.(3)The input-output relations of the deep learning model are discussed.According to the parameters used,the input parameters are arranged and combined and the influence factors are labeled according to the permutation and influence factor hypothesis method,and the input parameters that have a large influence on the measurement results in the deep learning model are identified.Then these input parameters were compared with the actual mathematical model and the results of robustness experiments.The connection between the deep learning model and the actual operation process is identified.A study related to soft measurement of some key thermal hydraulic parameters of reactor coolant systems in nuclear power plants is conducted.The results show that the proposed deep learning model has good performance in parameter soft measurement,and its error is smaller than that of the mechanistic model soft measurement method.Meanwhile,the robustness analysis proves that the deep learning model has certain robustness.Finally,this paper determines the correlation between the regulator water level and the one-loop flow rate with the evaporator water level through experiments.