Drive reinforcement neurals networks for reactor control. Final report

摘要

This research focused on the development and characterization of drive reinforcement neural networks for reactor power control. Drive reinforcement networks are a novel class of neural networks that have been shown to be capable of learning relatively complex tasks, such as reactor startup demonstrated here. The networks also have the capability to function as a supervisor of lower level controllers. This document briefly discusses the task of nuclear reactor power control and elements of control necessary for various types of controllers to be integrated into a hierarchical control system. Next, the application of a drive reinforcement network to multi-stage control is presented.