Neural-network accelerated coupled core-pedestal simulations with self-consistent transport of impurities and compatible with ITER IMAS

Abstract

An integrated modeling workflow capable of finding the steady-state plasma solution with self-consistent core transport, pedestal structure, current profile, and plasma equilibrium physics has been developed and tested against a DIII-D discharge. Key features of the achieved core-pedestal coupled workflow are its ability to account for the transport of impurities in the plasma self-consistently, as well as its use of machine learning accelerated models for the pedestal structure and for the turbulent transport physics. Notably, the coupled workflow is implemented within the One Modeling Framework for Integrated Tasks (OMFIT) framework, and makes use of the ITER integrated modeling and analysis suite data structure for exchanging data among the physics codes that are involved in the simulations. Such technical advance has been facilitated by the development of a new numerical library named ordered multidimensional arrays structure.

H-mode inhibition in negative triangularity tokamak reactor plasmas

“…This profile shape was chosen to provide a consistent and realistic Hmode edge profile which can be smoothly varied to an Lmode-like profile by controlling the pedestal height [39]. Further, this assumption facilitates integration with existing integrated modeling tools (like STEP) that make use of the EPED pedestal formalization [40]. To identify the critical gradient for infinite-n ballooning instability, the pedestal height in the EPED parameterization is scaled, keeping all other aspects of the profile constant, until a marginally-stable value of α is attained.…”

“…In this section, we apply the same concepts to a database of randomized equilibria in order to understand the robustness of the above trends, and to more fully characterize the broader reactor operational space. From this database, scaling laws for the normalized pressure gradient and the pedestal pressure can be generated, providing a simple starting point for the inclusion of physics dominating the NT edge into integrated tokamak models [40,43].…”

“…Alternatively, since infinite-n ballooning calculations are relatively straightforward and fast to run with codes like BALOO, it should be possible to extend this work in the future to produce a full database of equilibria with arbitrary profiles in order to construct a neural network for ballooning-limited profile prediction similar to EPEDNN. Such a task should be informed by experimental profiles as much as possible, and will be attempted after the additional experiments are performed on DIII-D. As part of the present engagement, a computational loop has been added to OMFIT's STEP framework [40,44] in order to calculate the ballooning boundary and iteratively adjust modeled profiles in the edge when relevant, allowing for profile predictions at δ < 0.…”

Reduced models for ETG transport in the pedestal

“…It utilizes a non-relational MongoDB [26] (NoSQL) based structure to maximize flexibility so that any output format from any particular code infrastructure can be easily supported. In addition, MGKDB seeks compatibility with an international IMAS data standard [27] to containerize its quantities of interest and interfaces these data types with a comprehensive python library called the Ordered Multidimensional Array Structures (OMAS) library [28], that allows for easy conversion to other data formats including, for example, SQL-based formats. MGKDB may be accessed remotely through either python scripts, command shell options (i.e.…”

EAST discharge prediction without integrating simulation results

“…An alternative approach for tokamak discharge modeling without relying on integrating the complex physical model is using the neural network method. The neural networks have been employed in magnetic fusion research to solve a variety of problems, including disruption prediction [4][5][6], simulation acceleration [7][8][9], plasma tomography [10], radiated power estimation [11], identifying instabilities [12], estimating neutral beam effects [13], classifying confinement regimes [14], determination of scaling laws [15,16], filament detection on MAST-U [17], electron temperature profile estimation via SXR with Thomson scattering [18], coil current prediction with the heat load pattern in W7-X [19], equilibrium reconstruction [18,[20][21][22][23][24], and equilibrium solver [25], control plasma [26][27][28][29][30][31], physic-informed machine learning [32]. Additionally, in our previous work [1], a neural-network-based method have been used on discharge modeling.…”