05/02494 Adaptive control of a PWR core power using neural networks

Abstract

This work proposes a new hybrid control strategy for a pressurized water type nuclear power plant by integrating linear quadratic integrator (LQI), linear quadratic Gaussian (LQG), and loop transfer recovery (LTR) approaches. The multi-input multi-output nuclear plant model adopted in this work is characterized by 38 state variables. The nonlinear plant model is linearized around steady-state operating conditions to obtain a linear model for the controller design. The proposed LQGI/LTR technique designs state-feedback assisted output control using the estimated states. The control architecture offers robust performance and tracks the reference set-point with zero steady-state error in the presence of uncertainties and disturbances. The effectiveness of the proposed technique is demonstrated by simulations on different subsections of a pressurized water nonlinear nuclear power plant model. The control performance of the proposed technique is further compared with other classical control design schemes. Statistical measures are employed to quantitatively analyse control performance.