A neural network controller for load following operation of nuclear reactors

Abstract

Nuclear reactors are in nature nonlinear and their parameters vary with time as a function of power level, fuel burnup, and control rod worth. Therefore, these characteristics must be considered if large power variations occur in power plant working regimes (for example in load following conditions). In this paper a neural network controller (NNC) is presented. A robust optimal self-tuning regulator (ROSTR) response is used as a reference trajectory to determine the feedback, feedforward and observer gains of the NNC. The NNC displayed good stability and performance for a wide range of operation as well as considerable reduction in computation time in regard to ROSTR and fuzzy logic controller (FAROC).