Identification and control of a nuclear reactor core (VVER) using recurrent neural networks and fuzzy systems

Abstract

Improving the methods of identification and control of nuclear power reactors core is an important area in nuclear engineering. Controlling the nuclear reactor core during load following operation encounters some difficulties in control of core thermal power while considering the core limitations in local power peaking and safety margins. In this paper, a nuclear power reactor core (VVER) is identified using a multi nonlinear autoregressive with exogenous inputs (NARX) structure, including neural networks with different time steps and a heuristic compound learning method, consisting of off- and on-line batch learning. An intelligent nuclear reactor core controller, is designed which possesses the fast data generation capabilities of the NARX neural network and a fuzzy system based on the operator knowledge and experience for the purpose of decision-making. The results of simulation with an accurate three-dimensional VVER core code show that the proposed controller is very well able to control the reactor core during load following operations, using optimum control rod group maneuver and variable overlapping strategy. This methodology represents an innovative method of core control using neuro-fuzzy systems and can be used for identification and control of other complex nonlinear plants.