Taking it to the next level: Managing complexity in the fusion plasma control system

Nuclear fusion plasmas require a plasma control system with a wide variety of functions and components for optimal and safe performance. A graph-based modelling framework tracks the integrated actuators and sensors, continuous plasma processes and variables, discrete plasma states and events, and requirements and defines the couplings between these. A Dependency Structure Matrix (DSM) analyses these couplings to reveal a potential global system layout.

The framework is demonstrated for ITER, resulting in a fully traceable graph model which suggests that the system can be organized into five distinct groups: Heating and current drive, magnetic configuration, burn dynamics, transport and exhaust, and plasma–wall interaction. All couplings between groups are made apparent in the DSM. Although ITER features specific actuators and sensors, these groups appear common for magnetically confined fusion devices.

The emerging structure highlights a structure of relatively independent control domains, that are weakly coupled. Model Predictive Control (MPC) has emerged as a strong candidate for plasma control in fusion devices owing to its ability to manage varying actuator and state constraints. We have derived MPCs for different domains like density, current density and temperature and exhaust control.

State-of-theart controllers are designed to operate within their designated domains, tracking their own reference signals without explicitly considering interactions with other controllers. Such a decentralized approach, is often suboptimal and can potentially fail to stabilize the overall interconnected system. Taking a systems level perspective, this strongly suggest a cooperative control strategy to account for the interaction between different domain-specific controllers. By extending the cost functions of the MPCs to incorporate interdomain coordination, we can achieve pareto optimal performance.