Development of PINN Controller for Fuel Handling System of Pressurized Heavy Water Reactors

Abstract

The required operating pressure of the Wolsong nuclear power plant is currently controlled by a proportional integral (PI) controller. The PI controller has a simple structure and was designed to meet requirements through gain setting. However, these control requirements can be difficult to meet without properly adjusting the gain when certain parameters change, such as the wear and tear in the valves or pipes. To solve these problems, it is important to dynamically change the PI controller gain or compensate for the PI controller output. The purpose of this study is to help design a controller that is capable of providing stable control in order to reduce errors regardless of parameter changes. The proposed PI neural network (PINN) control technique involves a PI controller and a neural network controller combined in parallel. The neural network component which is designed to be robust compensates the output of the controller for changes in the above-mentioned parameters. Because assessing the controller performance straightforwardly in real-time processes can be difficult, a simulator model was developed based on real-time processes, and it showed changes in the parameters involved. The results confirmed that the proposed PINN controller reduced the instability of the fuel supply machine and, hence the aforementioned problem could be properly controlled.