Magnetic control of tokamak plasmas through deep reinforcement learning

Nuclear fusion using magnetic confinement, in particular in the tokamak configuration, is a promising path towards sustainable energy. A core challenge is to shape and maintain a high-temperature plasma within the tokamak vessel. This requires high-dimensional, high-frequency, closed-loop control using magnetic actuator coils, further complicated by the diverse requirements across a wide range of plasma configurations. In this work, we introduce a previously undescribed architecture for tokamak magnetic controller design that autonomously learns to command the full set of control coils. This architecture meets control objectives specified at a high level, at the same time satisfying physical and operational constraints. This approach has unprecedented flexibility and generality in problem specification and yields a notable reduction in design effort to produce new plasma configurations. We successfully produce and control a diverse set of plasma configurations on the Tokamak à Configuration Variable1,2, including elongated, conventional shapes, as well as advanced configurations, such as negative triangularity and ‘snowflake’ configurations. Our approach achieves accurate tracking of the location, current and shape for these configurations. We also demonstrate sustained ‘droplets’ on TCV, in which two separate plasmas are maintained simultaneously within the vessel. This represents a notable advance for tokamak feedback control, showing the potential of reinforcement learning to accelerate research in the fusion domain, and is one of the most challenging real-world systems to which reinforcement learning has been applied.

Pedestal Characteristics and MHD Stability of H-Mode Plasmas in TCV

Andreas Pitzschke

The tokamak à configuration variable (TCV) is unique in its ability to create a variety of plasma shapes and to heat the electron population in high density regimes using microwave power at the third harmonic of the electron cyclotron frequency. In the frame of this thesis, the impact of plasma shaping and heating on the properties of the edge transport barrier (ETB) in the high confinement mode (H-mode) was studied. This mode of operation is foreseen as one of the reference scenarios for ITER, the International Tokamak Experimental Reactor, which is being built to demonstrate the feasibility of thermonuclear fusion using magnetic confinement. A feature of H-mode regime operation are edge localized modes (ELMs), instabilities driven by the steep pressure gradients that form in the plasma edge region due to a transport barrier. During an ELM event, energy and particles are expelled from the plasma in a short burst. This will cause serious problems with respect to the heat load on plasma facing components in a tokamak of the size of ITER. Understanding of the phenomena associated with ELMs is thus required and dedicated investigations of their theory and experimental observations are carried out in many laboratories worldwide. This thesis presents several experimental and numerical investigations of tokamak behavior for configurations where the plasma edge plays an important role. From the experimental viewpoint, studies of transport barriers are challenging, as plasma parameters change strongly within a narrow spatial region. As part of the work presented here, the TCV Thomson scattering system was upgraded to meet the requirements for diagnosing electron temperature and density with high spatial resolution in the region of internal and external transport barriers. Simultaneously, the data analysis was significantly improved to cope with statistical uncertainties and alleviate eventual systematic errors. For measurements of the time evolution of density and temperature profile during the ELM cycle, the low repetition rate of the lasers used for Thomson scattering is a limiting. Although the system on TCV comprises 3 laser units that may be triggered in sequence with time separations down to 1 ms, time evolution over longer periods can only be reconstructed from repetitive events. In this context, an adjustment of the laser trigger to improve the synchronization with the ELM event is an advantage. A method was developed and implemented to generate a synchronizing trigger sequence, by a real-time monitoring of the D-alpha emission, which provides a marker for the ELM event. Recently, a "snowflake" (SF) divertor configuration, proposed as a possible solution to reduce the plasma-wall interaction by changing the divertor's poloidal magnetic field topology, was generated, for the first time, in TCV. A numerical code (KINX), based on a magnetohydrodynamic model (ideal MHD), was used to investigate the stability limits of this configuration under H-mode conditions and compare them with a similar standard single-null equilibrium. In a series of experiments, improved energy confinement was found and explained by improved stability of the edge region in the SF configuration. The influence of the pedestal structure in ELMy H-mode plasmas on the energy confinement and on ELM energy losses was investigated. The different ELM regimes found in TCV were analyzed, in particular the transition between type-III to type-I ELMs. The operational boundary of each ELM regime was characterized and verified by ideal MHD stability simulations for the ETB region. Recent studies on the scaling of the pedestal width with normalized poloidal pressure were confirmed. Using the capabilities of TCV, the influence of plasma shaping on pedestal parameters and MHD stability limits was investigated. In the past, models were developed to describe the onset of type-I ELMs, which are associated with modes in the ETB region arising from a coupling of pressure- and current-driven instabilities (coupled kink-ballooning modes). Experimental studies were performed to trace the temporal evolution of pedestal parameters characterizing the ETB during an ELM cycle. The results of these experiments were analyzed using information from MHD stability calculations. It is concluded that these models are capable of predicting limits as necessary conditions for ELM activity, but are not sufficient to fully explain ELM triggering.

EPFL2011

Exploration of candidate fusion reactor regimes by real time control of tokamak plasma shape

Himank Anand

The potential of nuclear fusion to provide a practically inexhaustible source of energy has motivated scientists to work towards developing nuclear fusion tokamak power plants. Stable operation of a tokamak at high performance requires simultaneous treatment of several plasma control problems. Moreover, the complex physics which governs the tokamak plasma evolution must be studied and understood to make correct choices in controller design. This mutual inter-dependence has informed this thesis, using control solutions as an experimental tool for physics studies, and using physics knowledge for developing new advanced control solutions. The TCV tokamak at SPC-EPFL is ideally placed to explore issues at the interface between plasma physics and plasma control, by combining a state-of-the-art digital real time control system with a flexible and diverse set of actuators including a full set of independently powered shaping coils. The recent deployment of the real time version of the Grad-Shafranov equilibrium reconstruction code LIUQE, with a sub-ms cycle time in the digital control system, has facilitated the design of a new generalised plasma position and shape controller, based on the information on poloidal flux and magnetic field provided by the real-time Grad-Shafranov solver. The first issue addressed in the thesis is the development and experimental testing of a new real time control strategy to construct a generalised control algorithm for not only controlling the position of the plasma but also to aid in the precise control of higher order shape moments, X-points and strike points, particularly in advanced plasma configurations such as negative-triangularity plasmas, snowflake and super-X divertors, and doublets. A controller formulation ensuring flexibility through an ordering of controlled variables from the most easily to the least easily controlled, while respecting the hardware limits on the poloidal field coil currents, is developed. The successful experimental implementation of the control algorithm has been demonstrated for both fixed and time varying plasma position and shape for limiter and divertor plasma discharges. In addition, the controller has provided satisfactory performance with respect to plasma scenarios involving complex changes in the plasma shape and position. The second issue addressed in the thesis is the application of the generalised plasma position and shape controller to a snowflake plasma configuration. A comparison between the optimised generalised plasma position and shape controller with the performance of the TCV hybrid controller for a given reference snowflake plasma discharge showed a marked improvement in various geometrical properties of the snowflake plasma configuration in the vicinity of the null point. However, strong control of the poloidal magnetic field at the two X-points resulted in a tradeoff on the upper part of the plasma boundary, where the overall precision was comparable to that of the legacy controller. In the experimental time available, the snowflake shots developed exhibited a boundary that was too close to the inner wall of the vessel, modifying the edge plasma behaviour (studied with infrared cameras and Langmuir probes) and making it difficult to study the physics properties of the exact snowflake. However, further optimisation should well be possible.

EPFL2017

Physics Insight from Plasma Shaping of TCV Tokamak Plasmas - focus on Electron Heat Transport

Antoine Pochelon

In the search of a fusion reactor using magnetic confinement of toroidal plasmas, many important plasma performance parameters directly depend on the shape of the plasma cross-section. The unique shaping capability of the TCV tokamak ("Tokamak à Configuration Variable") has been exploited to study different aspects of tokamak transport and stability, in which the plasma shape may play a role. This presentation will focus on the effects of plasma shape on transport. The TCV tokamak has produced a host of physics results on diverse topics such as core sawtooth instabilities, edge localised modes (ELMs), electron cyclotron heating (ECH) and current drive (CD), plasma control, MHD stability, internal transport barriers (ITB), innovative “snowflake” divertors. TCV can produce plasmas with extreme shapes, elongation κ up to 2.8 and both negative and positive triangularity δ from –0.8 to +1, and is equipped with 4.5MW of localized Electron Cyclotron Heating (ECH). A significant fraction of TCV results has benefited from plasma shaping. The characterization of the various effects of shaping serves essentially two purposes: 1) it provides stringent tests for theory, being thus a valuable tool for model validation, and 2) it is crucial for the design of a fusion reactor, since plasma shape has a marked influence on confinement, MHD stability and performance. As an example, confinement in TCV has been demonstrated to depend strongly on triangularity in low-collisionality L-mode plasmas, improving towards negative triangularity. For illustration, the heat flux necessary to sustain the same profiles and stored energy in a discharge with δ=-0.4 is only half of that required for a discharge with δ=+0.4. The observed dependences of the electron thermal diffusivity on triangularity (direct) and collisionality (inverse) are qualitatively reproduced by nonlinear gyro-kinetic simulations and shown to be governed by Trapped Electron Mode (TEM) turbulence. Both in the linear and non-linear phases, negative triangularity is found to have a stabilizing influence on the TEM by modifying the toroidal precessional drift of trapped electrons that resonantly drive the modes. Negative triangularity also enhances the local shear, which in turn increases the perpendicular wave number, thus reducing the mixing length transport estimate. Both heat transport measurements and gyro-kinetic simulations demonstrate the stabilising effect of electron-ion collisions, which is now further sustained by recent electron temperature turbulence measurements using correlation-ECE.

2009