Session: 01-08: Nuclear Plant Operation, Modification, Life Extension, Maintenance and Life Cycle - VIII

Paper Number: 135988

135988 - A Control-Oriented Hybrid Model for Nuclear Reactors Based on Neural Network 

Abstract: 

Accurate modeling is the basis for analyzing the dynamic response characteristics of the model and exploring the actual physical processes. However, due to the complexity and time-varying nature of the internal mechanisms of the reactor, it is inevitable that inaccurate model parameters will be used in the modeling process, and some assumptions and simplifications will need to be made to the original complex model in order to consider the difficulty of the subsequent control system design work. These will lead to discrepancies between the modeled mechanisms and the actual reactor. In this paper, the difference is evaluated and shortened by means of neural network hybrid modeling. Considering that it is difficult to obtain the actual reactor operation data, this paper takes the data obtained through the mechanistic model as the actual operation data, and the linear model obtained by linearization based on the mechanistic model as the model to be corrected. In the actual operation process or the mechanism model, the physical parameters such as reactor temperature feedback coefficient, thermal conductivity of the metal pipe wall, specific heat capacity and so on change with the change of operating conditions, but the linear model is obtained based on a certain steady state operation operating point, and once obtained, it is kept unchanged during the transient operation process, so there is a deviation between the mechanism model and the linear model both in the transient operation process and in the final stabilized results, which might cause the control system designed based on the linear model to be difficult to achieve the desired control effect in the mechanistic model, and the error can be reduced by using neural network to correct the linear model. Based on the MATLAB/Simulink simulation platform, this paper firstly obtains the parameters that have the greatest influence on the linear model through sensitivity analysis and takes them as the object of neural network correction, then obtains the data required for offline training of neural network according to the mechanism model, linear model and the deviation of the two under different working condition levels, retains the neural network weights and thresholds obtained from the offline training, and finally utilizes the gradient descent algorithm to update the neural network weights and thresholds in real time in order to achieve the on-line calibration of the linear model. The final results show that after the neural network correction, the hybrid model can better correct the transient overshoot, regulation time and other key control system parameters of interest and effectively reduce the steady-state deviation between the linear model and the mechanistic model, which indicates that the hybrid modeling by means of neural network can effectively improve the accuracy of the established model and provide a solid foundation for the subsequent design of the control system based on the linear model.

Key works: Hybrid modeling; Mechanistic model; Linear model; Neural network; Gradient descent algorithm

Presenting Author: Ze Zhu Xi'an Jiaotong University

Presenting Author Biography: My name is Ze Zhu and I am a graduate student at Xi'an Jiaotong University. My school is located in Xi'an, Shaanxi Province, China, a city famous for its long history and unique culture. My major is nuclear science and technology, and my research focus is on modeling and control algorithms for nuclear reactor systems. At present, I have established models of some key equipments such as reactors and steam generators based on the three conservation equations and nuclear reactor dynamics. I have used PID, sliding mode control, and predictive control to design the reactor power control system, average temperature control system, and steam generator secondary side pressure control system, and have achieved good control results.

Authors:

Ze Zhu Xi'an Jiaotong University
Wenlong Liang Xi'an Jiaotong University
Baiqing Ye Xi'an Jiaotong University
Qingfeng Jiang Xi'an Jiaotong University
Pengfei Wang Xi'an Jiaotong University