Self-consistent core-pedestal transport simulations with neural network accelerated models

Meneghini, O.; Smith, S. P.; Snyder, P. B.; Staebler, G. M.; Candy, J.; Belli, E. A.; Lao, L. L.; Kostuk, Mark; Luce, T. C.; Luda, T.; Park, J.M.; Poli, F. M.

doi:10.1088/1741-4326/aa7776

Cited by 96 publications

(98 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Further work on improving and extending the RAPTOR code for predictive simulation purposes will focus on coupling to free-and fixed-boundary Grad-Shafranov equlibrium solvers, as well as on fast models to estimate the sources and sinks of power and particles (including radiation). Furthermore, use of core-pedestal models is envisaged, for example using a neural network regression of pedestal heights and widths derived from the EPED model [45], as in [20]. Coupling to the edge would require validated, reduced models of the plasma in that region (e.g.…”

Section: Discussionmentioning

confidence: 99%

“…Note that while it is known that in reality H 98 tends to increase with power at JET [44], (owing an the overly pessimistic P −0.7 term in the H 98 scaling), the purpose of this model is to demonstrate the core-pedestal coupling in these time-dependent simulations. In the future, this capability can be used to include a more physics-based pedestal model into these simulations, for example as in [20].…”

Section: Nbi Input Power Scan Including Core-pedestal Couplingmentioning

confidence: 99%

See 1 more Smart Citation

Real-time-capable prediction of temperature and density profiles in a tokamak using RAPTOR and a first-principle-based transport model

et al. 2018

View full text Add to dashboard Cite

The RAPTOR code is a control-oriented core plasma profile simulator with various applications in control design and verification, discharge optimization and realtime plasma simulation. To date, RAPTOR was capable of simulating the evolution of poloidal flux and electron temperature using empirical transport models, and required the user to input assumptions on the other profiles and plasma parameters. We present an extension of the code to simulate the temperature evolution of both ions and electrons, as well as the particle density transport. A proof-of-principle neural-network emulation of the quasilinear gyrokinetic QuaLiKiz transport model is coupled to RAPTOR for the calculation of first-principle-based heat and particle turbulent transport. These extended capabilities are demonstrated in a simulation of a JET discharge. The multi-channel simulation requires ∼ 0.2 second to simulate 1 second of a JET plasma, corresponding to ∼20 energy confinement times, while predicting experimental profiles within the limits of the transport model. The transport model requires no external inputs except for the boundary condition at the top of the H-mode pedestal. This marks the first time that simultaneous, accurate predictions of Te, T i and ne have been obtained using a first-principle-based transport code that can run in faster-than-real-time for present-day tokamaks.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Nbi Input Power Scan Including Core-pedestal Couplingmentioning

confidence: 99%

Real-time-capable prediction of temperature and density profiles in a tokamak using RAPTOR and a first-principle-based transport model

et al. 2018

View full text Add to dashboard Cite

show abstract

“…For example, ‘trained’ transport models based on neural network techniques have been developed using the reduced transport models TGLF (Meneghini et al. 2017) and QuaLiKiz (Citrin et al. 2015), using compiled databases of transport model outputs.…”

Section: Frontiers In Gyrokinetic Transport Model Validationmentioning

confidence: 99%

Validation of nonlinear gyrokinetic transport models using turbulence measurements

White

2019

J. Plasma Phys.

View full text Add to dashboard Cite

This tutorial covers validation of gyrokinetic turbulent-transport models via comparison of measured turbulence with realistic simulations of fusion plasmas. It presents a brief history of validation of gyrokinetic simulations, the principal challenges encountered, a limited survey of common turbulence diagnostics used on tokamaks and stellarators, an overview of the fundamentals of synthetic diagnostic models and a discussion of frontiers in turbulent-transport model validation.

show abstract

“…As alluded to in Section 5, in coupling turbulence simulations with transport, it is natural for the flux that is passed from the turbulence simulation to the transport solver to be averaged over a time T to yield q , as in Eq. (8). We use q as the heat flux at each l iteration of Eqs.…”

Section: Non-gaussian Noise: Temporally Correlated Spatially Uniformmentioning

confidence: 99%

Investigation of a Multiple-Timescale Turbulence-Transport Coupling Method in the Presence of Random Fluctuations

Parker

LoDestro

Campos

2018

Plasma

View full text Add to dashboard Cite

One route to improved predictive modeling of magnetically confined fusion reactors is to couple transport solvers with direct numerical simulations (DNS) of turbulence, rather than with surrogate models. An additional challenge presented by coupling directly with DNS is that the inherent fluctuations in the turbulence, which limit the convergence achievable in the transport solver. In this article, we investigate the performance of one numerical coupling method in the presence of turbulent fluctuations. To test a particular numerical coupling method for the transport solver, we use an autoregressive-moving-average model to efficiently generate stochastic fluctuations with statistical properties resembling those of a gyrokinetic simulation. These fluctuations are then added to a simple, solvable problem, and we examine the behavior of the coupling method. We find that monitoring the residual as a proxy for the error can be misleading. From a pragmatic point of view, this study aids us in the full problem of transport coupled to DNS by predicting the amount of averaging required to reduce the fluctuation error and obtain a specific level of accuracy.

show abstract

Self-consistent core-pedestal transport simulations with neural network accelerated models

Cited by 96 publications

References 42 publications

Real-time-capable prediction of temperature and density profiles in a tokamak using RAPTOR and a first-principle-based transport model

Real-time-capable prediction of temperature and density profiles in a tokamak using RAPTOR and a first-principle-based transport model

Validation of nonlinear gyrokinetic transport models using turbulence measurements

Investigation of a Multiple-Timescale Turbulence-Transport Coupling Method in the Presence of Random Fluctuations

Contact Info

Product

Resources

About