THE DEVELOPMENT OF MODEL PREDICTIVE CONTROL FOR NUCLEAR REACTOR

Abstract

Different power levels of a nuclear reactor have different behavior due to nonlinearity and time variance parameters. So, the need for a controller to handle these constraints and work at different power levels is essential. The work in this thesis provides a design of two controllers and apply these controllers over a multi-point reactor core model. This model has sufficient accuracy to represent the reactor and has the main neutron reaction and the six precursors delayed neutron. The effect of temperature changes and the nonlinearity of Xenon poisoning have been taken into consideration in designing this model. This model also takes into consideration the effect of fuel consumption during the operation. In this thesis, two controllers have been designed. These controllers are Model Predictive control and Neural Network Predictive control. These controllers are tested for tracking and regulating scenarios at different power levels. These scenarios are evaluated for enough time to test the nonlinearity behavior and time variance of the system and their results are compared to the currently installed Proportional Derivative controller. Although filters are applied over the sensor system for both neutron noise and signal noise, the controllers are tested for nonfiltered signals. The model predictive controller has shown its performance superiority over the other controllers. It handles system constraints and dynamics. It has the least variation over the steady state, no overshoot, no steady state error and with the fastest action.