Dynamic Mode Decomposition For The Control Of Nuclear Power Plants

The control of high-dimensional systems is a challenging task, mostly because many control approaches successfully implemented and used in small systems do not scale well as the domain dimension increase. Controllers for full, industrial systems may be computationally prohibitive, and thus cannot be used for real-time monitoring. For this reason, practical control strategies for dealing with high-dimensional data employs reduced order methods to design low-dimensional sub-spaces where computationally feasible controllers can be used in real-time applications. A new technique developed in this framework is dynamic mode decomposition with control (DMDc). Based upon the data-driven, equation-free DMD algorithm, which reconstruct the underlying dynamics of the system from snapshot measurements, DMDc is particularly suited for systems with nonlinear dynamics, such as nuclear power plants. In this work, the DMDc algorithm is tweaked and applied to the study of the stability of a Pressurized Water Reactor (PWR) and its response to known inputs, as well as to understand how control affects the system. The results show that the DMDc is able to provide accurate input-output models for complex systems with external forcing, without the need to know the underlying dynamics beforehand and with reasonable computational efforts, also suitable for real-time applications.