skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Design and optimization of Artificial Neural Networks for the modelling of superconducting magnets operation in tokamak fusion reactors

Abstract

In superconducting tokamaks, the cryoplant provides the helium needed to cool different clients, among which by far the most important one is the superconducting magnet system. The evaluation of the transient heat load from the magnets to the cryoplant is fundamental for the design of the latter and the assessment of suitable strategies to smooth the heat load pulses, induced by the intrinsically pulsed plasma scenarios characteristic of today's tokamaks, is crucial for both suitable sizing and stable operation of the cryoplant. For that evaluation, accurate but expensive system-level models, as implemented in e.g. the validated state-of-the-art 4C code, were developed in the past, including both the magnets and the respective external cryogenic cooling circuits. Here we show how these models can be successfully substituted with cheaper ones, where the magnets are described by suitably trained Artificial Neural Networks (ANNs) for the evaluation of the heat load to the cryoplant. First, two simplified thermal-hydraulic models for an ITER Toroidal Field (TF) magnet and for the ITER Central Solenoid (CS) are developed, based on ANNs, and a detailed analysis of the chosen networks' topology and parameters is presented and discussed. The ANNs are then inserted into the 4C model of themore » ITER TF and CS cooling circuits, which also includes active controls to achieve a smoothing of the variation of the heat load to the cryoplant. The training of the ANNs is achieved using the results of full 4C simulations (including detailed models of the magnets) for conventional sigmoid-like waveforms of the drivers and the predictive capabilities of the ANN-based models in the case of actual ITER operating scenarios are demonstrated by comparison with the results of full 4C runs, both with and without active smoothing, in terms of both accuracy and computational time. Exploiting the low computational effort requested by the ANN-based models, a demonstrative optimization study has been finally carried out, with the aim of choosing among different smoothing strategies for the standard ITER plasma operation.« less

Authors:
; ; ; ; ;
Publication Date:
OSTI Identifier:
22572350
Resource Type:
Journal Article
Resource Relation:
Journal Name: Journal of Computational Physics; Journal Volume: 321; Other Information: Copyright (c) 2016 Elsevier Science B.V., Amsterdam, The Netherlands, All rights reserved.; Country of input: International Atomic Energy Agency (IAEA)
Country of Publication:
United States
Language:
English
Subject:
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSICS; ACCURACY; DESIGN; HEATING LOAD; ITER TOKAMAK; NEURAL NETWORKS; OPTIMIZATION; PLASMA; PULSES; SIMULATION; SOLENOIDS; SUPERCONDUCTING MAGNETS; THERMAL HYDRAULICS; WAVE FORMS

Citation Formats

Froio, A., Bonifetto, R., Carli, S., Quartararo, A., Savoldi, L., E-mail: laura.savoldi@polito.it, and Zanino, R. Design and optimization of Artificial Neural Networks for the modelling of superconducting magnets operation in tokamak fusion reactors. United States: N. p., 2016. Web. doi:10.1016/J.JCP.2016.05.028.
Froio, A., Bonifetto, R., Carli, S., Quartararo, A., Savoldi, L., E-mail: laura.savoldi@polito.it, & Zanino, R. Design and optimization of Artificial Neural Networks for the modelling of superconducting magnets operation in tokamak fusion reactors. United States. doi:10.1016/J.JCP.2016.05.028.
Froio, A., Bonifetto, R., Carli, S., Quartararo, A., Savoldi, L., E-mail: laura.savoldi@polito.it, and Zanino, R. Thu . "Design and optimization of Artificial Neural Networks for the modelling of superconducting magnets operation in tokamak fusion reactors". United States. doi:10.1016/J.JCP.2016.05.028.
@article{osti_22572350,
title = {Design and optimization of Artificial Neural Networks for the modelling of superconducting magnets operation in tokamak fusion reactors},
author = {Froio, A. and Bonifetto, R. and Carli, S. and Quartararo, A. and Savoldi, L., E-mail: laura.savoldi@polito.it and Zanino, R.},
abstractNote = {In superconducting tokamaks, the cryoplant provides the helium needed to cool different clients, among which by far the most important one is the superconducting magnet system. The evaluation of the transient heat load from the magnets to the cryoplant is fundamental for the design of the latter and the assessment of suitable strategies to smooth the heat load pulses, induced by the intrinsically pulsed plasma scenarios characteristic of today's tokamaks, is crucial for both suitable sizing and stable operation of the cryoplant. For that evaluation, accurate but expensive system-level models, as implemented in e.g. the validated state-of-the-art 4C code, were developed in the past, including both the magnets and the respective external cryogenic cooling circuits. Here we show how these models can be successfully substituted with cheaper ones, where the magnets are described by suitably trained Artificial Neural Networks (ANNs) for the evaluation of the heat load to the cryoplant. First, two simplified thermal-hydraulic models for an ITER Toroidal Field (TF) magnet and for the ITER Central Solenoid (CS) are developed, based on ANNs, and a detailed analysis of the chosen networks' topology and parameters is presented and discussed. The ANNs are then inserted into the 4C model of the ITER TF and CS cooling circuits, which also includes active controls to achieve a smoothing of the variation of the heat load to the cryoplant. The training of the ANNs is achieved using the results of full 4C simulations (including detailed models of the magnets) for conventional sigmoid-like waveforms of the drivers and the predictive capabilities of the ANN-based models in the case of actual ITER operating scenarios are demonstrated by comparison with the results of full 4C runs, both with and without active smoothing, in terms of both accuracy and computational time. Exploiting the low computational effort requested by the ANN-based models, a demonstrative optimization study has been finally carried out, with the aim of choosing among different smoothing strategies for the standard ITER plasma operation.},
doi = {10.1016/J.JCP.2016.05.028},
journal = {Journal of Computational Physics},
number = ,
volume = 321,
place = {United States},
year = {Thu Sep 15 00:00:00 EDT 2016},
month = {Thu Sep 15 00:00:00 EDT 2016}
}
  • In this paper, we examine the possibility of using a multilayered feedforward neural network to extract tokamak plasma parameters from magnetic measurements as an improvement over the traditional methodology of function parametrization. It is also used to optimize the number and locations of the magnetic diagnostics designed for the tokamak. This work has been undertaken with the specific purpose of application of the neural network technique to the newly designed (and currently under fabrication) Superconducting Steady-State Tokamak-1 (SST-1). The magnetic measurements will be utilized to achieve real-time control of plasma shape, position, and some global profiles. A trained neural networkmore » is tested, and the results of parameter identification are compared with function parametrization. Both techniques appear well suited for the purpose, but a definite improvement with neural networks is observed. Although simulated measurements are used in this work, confidence regarding the network performance with actual experimental data is ensured by testing the network's noise tolerance with Gaussian noise of up to 10%. Finally, three possible methods of ranking the diagnostics in decreasing order of importance are suggested, and the neural network is used to optimize the number and locations of the magnetic sensors designed for SST-1. The results from the three methods are compared with one another and also with function parametrization. Magnetic probes within the plasma-facing side of the outboard limiter have been ranked high. Function parametrization and one of the neural network methods show a distinct tendency to favor the probes in the remote regions of the vacuum vessel, proving the importance of redundancy. Fault tolerance of the optimized network is tested. The results obtained should, in the long run, help in the decision regarding the final effective set of magnetic diagnostics to be used in SST-1 for reconstruction of the control parameters.« less
  • An optimization artificial neural networks model is developed for solving the ill-posed inverse transport problem associated with localizing radioactive sources in a medium with known properties and dimensions. The model is based on the recurrent (or feedback) Hopfield network with fixed weights. The source distribution is determined based on the response of a limited number of external detectors of known spatial deployment in conjunction with a radiation transport model. The algorithm is tested and evaluated for a large number of simulated two-dimensional cases. Computations are carried out at different noise levels to account for statistical errors encountered in engineering applications.more » The sensitivity to noise is found to depend on the number of detectors and on their spatial deployment. A pretest empirical procedure is, therefore, suggested for determining an effective arrangement of detectors for a given problem.« less
  • A new approach to nonlinear groundwater management methodology is presented which optimizes aquifer remediation with the aid of artificial neural networks (ANNs). The methodology allows solute transport simulations, usually the main computational component of management models, to be run in parallel. The ANN technology, inspired by neurobiological theories of massive interconnection and parallelism, has been successfully applied to a variety of optimization problems. In this new approach, optimal management solutions are found by (1) first training an ANN to predict the outcome of the flow and transport code, and (2) then using the trained ANN to search through many pumpingmore » realizations to find an optimal one for successful remediation. The behavior of complex groundwater scenarios with spatially variable transport parameters and multiple contaminant plumes is simulated with a two-dimensional hybrid finite-difference/finite-element flow and transport code. The flow and transport code develops the set of examples upon which the network is trained. The input of the ANN characterizes the different realizations of pumping, with each input indicating the pumping level of a well. The output is capable of characterizing the objectives and constraints of the optimization, such as attainment of regulatory goals, value of cost functions and cleanup time, and mass of contaminant removal. The supervised learning algorithm of back propagation was used to train the network. The conjugate gradient method and weight elimination procedures are used to speed convergence and improve performance, respectively. Once trained, the ANN begins a search through various realizations of pumping patterns to determine whether or not they will be successful. The search is directed by a simple genetic algorithm. The resulting management solutions are consistent with those resulting from a more conventional optimization technique. 84 refs., 9 figs., 5 tabs.« less
  • In a long-haul sensor network, sensors are remotely deployed over a large geographical area to perform certain tasks, such as tracking and/or monitoring of one or more dynamic targets. A remote fusion center fuses the information provided by these sensors so that a final estimate of certain target characteristics – such as the position – is expected to possess much improved quality. In this paper, we pursue learning-based approaches for estimation and fusion of target states in longhaul sensor networks. In particular, we consider learning based on various implementations of artificial neural networks (ANNs). Finally, the joint effect of (i)more » imperfect communication condition, namely, link-level loss and delay, and (ii) computation constraints, in the form of low-quality sensor estimates, on ANN-based estimation and fusion, is investigated by means of analytical and simulation studies.« less