A general infrastructure for data-driven control design and implementation in tokamaks. (17th February 2023)

• Record Type: Journal Article
• Authors: Abbate, Joseph; Conlin, Rory; Shousha, Ricardo; Erickson, Keith; Kolemen, Egemen
• Is Part Of: Journal of plasma physics. Volume 89:Number 1(2023)
• Publication Date: 2023-02-17
• Subjects: fusion plasma -- plasma dynamics; Plasma (Ionized gases) -- Periodicals; 530.4405
• DOI: 10.1017/S0022377822001040
• Languages: English
• ISSNs: 0022-3778

Abstract

Abstract : A general infrastructure for tokamak controllers based on data-driven neural net models is presented. The paradigm allows for more flexible choices of both the underlying model and the desired controlled variables and targets. The system is implemented and tested on the DIII-D tokamak, enacting simultaneous pressure and temperature control via a finite-set model-predictive controller. Traditional control methods such as proportional–integral–derivative (PID) have proven effective for decoupled control tasks, but scale poorly when trying to achieve more complicated goals such as full state control. This is exactly where model-based controllers succeed.