Research on Hybrid Modeling Method Based on Mechanism Model and BP Neural Network

Reactor system modeling is the premise of reactor control system design. The common commercial thermal and hydraulic analysis programs like RELAP5, RETRAN have complex structure and slow operational speed, which are not conductive to the transplantation and improvement of reactor system. In order to realize the rapid development of reactor control system, this paper carries out the mechanism modeling of the reactor system in Matlab/Simulink, and then introduce BP neural network to establish the hybrid model, which improves the modeling accuracy to support the design of reactor system. For small pressurized water reactor, we firstly take analytical modeling method to build the core mechanism model and its state space model through three conservation principles and the operating mechanism of core. Based on analyses, we can conclude that the parameter error and structural error of the mechanism model is caused by the assumption and simplification during modeling, and choose the parameter with the most influence (the feedback coefficient of the coolant temperature) as the corrected parameter in the BP neural-based hybrid modeling process. After that, according to the dual constant operation scheme of small pressurized water reactors, and the gradient descent method of BP network, the hybrid model of core is built and the parameter error and structure error of the mechanism modeling are corrected online. The simulation results show that the proposed hybrid modeling method can accurately simulate the reactor operate state. Thus, the proposed hybrid model is accurate and effective.