Dynamic Neural Network-based Feedback Linearization Control of a Pressurized Water Reactor

Abstract

This note presents a nonlinear control approach using dynamic neural network (DNN)-based feedback linearization (FBL) for nuclear reactor power control. The reactor model adopted in this study is based on neutronic dynamic and thermal-hydraulic models. The nonlinear plant is identified by a single-layer DNN trained using Quasi-Newton and Interior-Point methods. The feedback linearization scheme is combined with a Proportional-Integral (P-I) controller and simulations show good performance of the proposed controller. The efficacy of the controller is evaluated in the load-following mode of operation. Moreover, the fault-tolerance performance of the proposed approach is tested.