Modeling of the HL-2A plasma vertical displacement control system based on deep learning and its controller design

Abstract

The modeling and control of the plasma equilibrium response is still one of the more important research areas in tokamak discharge experiments. Although theoretically, first principles can predict the plasma instability, how to build a physical model for accurate prediction is still a challenging problem. Therefore, a deep learning method is proposed to model the plasma vertical displacement system in the HL-2A tokamak experiment, whose method expands the modeling strategy for tokamak plasma control systems. Through the training of a large number of high-dimensional experimental data, the obtained deep neural network model in this paper has a higher precision prediction ability. Additionally, to illustrate the significance of the predictive model in controller design, a data-driven adaptive control algorithm is proposed to replace the traditional proportional-integral-derivative control algorithm for controlling the vertical displacement of plasma. The simulation results showed that the proposed algorithm had less adjustable parameters, strong self-adaptability, and effective control for the HL-2A plasma vertical displacement.

Experiment data-driven modeling of tokamak discharge in EAST

“…The history can be traced back to the use of machine learning for disruption prediction since the 1990s, i.e. ADITYA [11,12], Alcator C-Mod [13][14][15], EAST [15][16][17], DIII-D [13,15,16,[18][19][20][21][22][23][24][25][26][27][28], JET [16,22,[28][29][30][31][32][33][34][35][36], ASDEX-Upgrade [30,[37][38][39], JT-60U [40][41][42], HL-2A [43,44], NSTX [45], and J-TEXT [46,47]. We note that applications of neural networks in the fusion community are increasing rapidly, and examples are simulation acceleration [48][49][50], plasma tomography …”

“…We note that applications of neural networks in the fusion community are increasing rapidly, and examples are simulation acceleration [48][49][50], plasma tomography [36], radiated power estimation [51], identifying instabilities [52], estimating neutral beam effects [53], classifying confinement regimes [54], determination of scaling laws [55,56], filament detection on MAST-U [57], electron temperature profile estimation via SXR with Thomson scattering [58], coil current prediction with the heat load pattern in W7-X [59], equilibrium reconstruction [58,[60][61][62][63][64], and equilibrium solver [65]. Additionally, machine learning methods have been tested for control purposes, in some cases to accelerate the computation [44,[66][67][68][69]. The Bishop [66] work is the first use of neural networks for real-time feedback control in a tokamak fusion experiment.…”

Configuration characteristics of the Chinese First Quasi-axisymmetric Stellarator

“…With respect to further optimizations of coils we found that the surface torsions of finite-sized coils are greatly influential on the simplification of modular coils and fabrication accuracy. We have figured out how to design a surface-torsion-free coil system and a set of practical coils without surface torsions was accomplished for the CFQS [26]. The surface torsions merely exist in finite-sized coils rather than filament coils.…”

EAST discharge prediction without integrating simulation results

“…A neural network based method is an alternative approach for tokamak discharge prediction without integrating complex physical modeling. The method has been adopted in magnetic fusion research to solve a variety of problems, including disruption prediction [3][4][5][6][7][8][9], electron temperature profile estimation from multi-energy SXR diagnostics [10], radiated power estimation [11], filament detection [12], simulation acceleration [13][14][15], classifying confinement regimes [16], plasma tomography [17], identification of instabilities [18], estimation of neutral beam effects [19], determination of scaling laws [20,21] coil current prediction with the heat load pattern [22], equilibrium reconstruction [10,[23][24][25][26][27], and equilibrium solver [28], control plasma [29][30][31][32][33][34], physicinformed machine learning [35]. Additionally, a method mixed neural-network with simulation code for discharge prediction [36] was noted.…”