Next Step Fusion Simulator
The Plasma Control System (PCS) for next-generation tokamaks and future fusion reactors is a critical component that governs plasma behavior, ensuring both stability and reliability. To meet the stringent requirements of regulators, the PCS must be more than just high-quality software—it must integrate state-of-the-art technologies and solutions that provide dependable control, even under challenging conditions such as long pulses, limited diagnostics, weak actuators, drifting signals, and more.
Plasma’s chaotic nature makes it inherently difficult to control. At Next Step Fusion, we believe the key to effective plasma management lies in combining conventional control methods, such as PID controllers, with machine learning (ML)-based techniques. This hybrid approach ensures that control systems not only maintain plasma performance but also enhance the efficiency of research and operations.
We develop modern ML-based Plasma Control System (PCS) and specialized control solutions for current and next-generation tokamaks and future fusion reactors. For example, in 2023-24, as participants in the DIII-D User Program, we experimented with and applied reinforcement learning (RL) algorithms to train ML models for plasma parameter control, as well as for training an ML model for plasma shape reconstruction. These efforts paved the way for building critical components of the PCS.
Plasma’s chaotic nature makes it inherently difficult to control. At Next Step Fusion, we believe the key to effective plasma management lies in combining conventional control methods, such as PID controllers, with machine learning (ML)-based techniques. This hybrid approach ensures that control systems not only maintain plasma performance but also enhance the efficiency of research and operations.
We develop modern ML-based Plasma Control System (PCS) and specialized control solutions for current and next-generation tokamaks and future fusion reactors. For example, in 2023-24, as participants in the DIII-D User Program, we experimented with and applied reinforcement learning (RL) algorithms to train ML models for plasma parameter control, as well as for training an ML model for plasma shape reconstruction. These efforts paved the way for building critical components of the PCS.
-
Novel Control Approach
Our PCS is based on a novel machine-agnostic and plasma-state-oriented control approach, which includes a machine-agnostic layer with a Plasma State Monitor (PSM) and Supervisory Controller (SC). In this context, “machine-agnostic” means that it can represent the plasma state of any magnetic confinement machine (tokamak or stellarator). This is achieved by unifying plasma states using mathematical models common to all machines. -
ML-enabled Design
The PCS employs both conventional and ML algorithms to ensure key components operate at the required levels of performance, reliability, and stability. For example, the PSM, an advanced classifier, integrates traditional and ML models with data from the Plasma State Reconstructor and real-time simulations from the Flight Simulator to provide detailed plasma state representation and support core-edge modeling through a unified interface. -
High-quality Software
Our goal is to provide the industry with next-level technology readiness (TRL) high-quality software. Unlike previous generations of scientific codes, the principles behind our PCS design and development are guided by best practices in software development, machine learning, fusion, and the nuclear industry’s standards for energy-producing reactors.
