EFIT-Prime: Probabilistic and physics-constrained reduced-order neural network model for equilibrium reconstruction in DIII-D

We introduce EFIT-Prime, a novel machine learning surrogate model for EFIT (Equilibrium FIT) that integrates probabilistic and physics-informed methodologies to overcome typical limitations associated with deterministic and ad hoc neural network architectures. EFIT-Prime utilizes a neural architectu...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Physics of plasmas 2024-09, Vol.31 (9)
Hauptverfasser: Madireddy, S., Akçay, C., Kruger, S. E., Amara, T. Bechtel, Sun, X., McClenaghan, J., Koo, J., Samaddar, A., Liu, Y., Balaprakash, P., Lao, L. L.
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue 9
container_start_page
container_title Physics of plasmas
container_volume 31
creator Madireddy, S.
Akçay, C.
Kruger, S. E.
Amara, T. Bechtel
Sun, X.
McClenaghan, J.
Koo, J.
Samaddar, A.
Liu, Y.
Balaprakash, P.
Lao, L. L.
description We introduce EFIT-Prime, a novel machine learning surrogate model for EFIT (Equilibrium FIT) that integrates probabilistic and physics-informed methodologies to overcome typical limitations associated with deterministic and ad hoc neural network architectures. EFIT-Prime utilizes a neural architecture search-based deep ensemble for robust uncertainty quantification, providing scalable and efficient neural architectures that comprehensively quantify both data and model uncertainties. Physically informed by the Grad–Shafranov equation, EFIT-Prime applies a constraint on the current density Jtor and a smoothness constraint on the first derivative of the poloidal flux, ensuring physically plausible solutions. Furthermore, the spatial location of the diagnostics is explicitly incorporated in the inputs to account for their spatial correlation. Extensive evaluations demonstrate EFIT-Prime's accuracy and robustness across diverse scenarios, most notably showing good generalization on negative-triangularity discharges that were excluded from training. Timing studies indicate an ensemble inference time of 15 ms for predicting a new equilibrium, offering the possibility of plasma control in real-time, if the model is optimized for speed.
doi_str_mv 10.1063/5.0213609
format Article
fullrecord <record><control><sourceid>proquest_osti_</sourceid><recordid>TN_cdi_osti_scitechconnect_2477694</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>3101349878</sourcerecordid><originalsourceid>FETCH-LOGICAL-c209t-5dc463557c438e0c39a2bd8b24071939bf074f60899abf70110ebaa8e123d34b3</originalsourceid><addsrcrecordid>eNp9kUtLxTAQhYso-Fz4D4KuFKKTJk0ad-KzIOhCwV3Iqxi9t7kmLeLC_24ude1qZvHNmTlzquqQwBkBTs-bM6gJ5SA3qh0CrcSCC7a57gVgztnrdrWb8zsAMN60O9XPzW33jJ9SWPoL9JSi0SYsQh6DRXpwaPX2nYPN2MYhj0mHwTuUvJusdzgm5xMa_JT0opTxK6YPtIzOL1AfE_KfU1EyKUzLMjILTHYMcUBhQNdd1-Hr_Wqr14vsD_7qXvVye_N8dY8fHu-6q8sHbGuQI26cZZw2jbCMth4slbo2rjU1A0EklaYHwXpe7EptegGEgDdat57U1FFm6F51NOvG4kxlG0Zv38pJg7ejqpkQXLICHc_QKsXPyedRvccpDeUuRQkQymQr2kKdzJRNMefke7Uqz9PpWxFQ6whUo_4iKOzpzK436rX1f-BftuuGJQ</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>3101349878</pqid></control><display><type>article</type><title>EFIT-Prime: Probabilistic and physics-constrained reduced-order neural network model for equilibrium reconstruction in DIII-D</title><source>Alma/SFX Local Collection</source><creator>Madireddy, S. ; Akçay, C. ; Kruger, S. E. ; Amara, T. Bechtel ; Sun, X. ; McClenaghan, J. ; Koo, J. ; Samaddar, A. ; Liu, Y. ; Balaprakash, P. ; Lao, L. L.</creator><creatorcontrib>Madireddy, S. ; Akçay, C. ; Kruger, S. E. ; Amara, T. Bechtel ; Sun, X. ; McClenaghan, J. ; Koo, J. ; Samaddar, A. ; Liu, Y. ; Balaprakash, P. ; Lao, L. L. ; General Atomics, San Diego, CA (United States)</creatorcontrib><description>We introduce EFIT-Prime, a novel machine learning surrogate model for EFIT (Equilibrium FIT) that integrates probabilistic and physics-informed methodologies to overcome typical limitations associated with deterministic and ad hoc neural network architectures. EFIT-Prime utilizes a neural architecture search-based deep ensemble for robust uncertainty quantification, providing scalable and efficient neural architectures that comprehensively quantify both data and model uncertainties. Physically informed by the Grad–Shafranov equation, EFIT-Prime applies a constraint on the current density Jtor and a smoothness constraint on the first derivative of the poloidal flux, ensuring physically plausible solutions. Furthermore, the spatial location of the diagnostics is explicitly incorporated in the inputs to account for their spatial correlation. Extensive evaluations demonstrate EFIT-Prime's accuracy and robustness across diverse scenarios, most notably showing good generalization on negative-triangularity discharges that were excluded from training. Timing studies indicate an ensemble inference time of 15 ms for predicting a new equilibrium, offering the possibility of plasma control in real-time, if the model is optimized for speed.</description><identifier>ISSN: 1070-664X</identifier><identifier>EISSN: 1089-7674</identifier><identifier>DOI: 10.1063/5.0213609</identifier><identifier>CODEN: PHPAEN</identifier><language>eng</language><publisher>Melville: American Institute of Physics</publisher><subject>Algorithms and data structure ; Artificial neural networks ; Bayesian statistics ; Constraints ; Equilibrium ; Fusion energy ; Machine learning ; Magnetohydrodynamics ; Neural networks ; Nuclear fusion ; Plasma control ; Poloidal flux ; Predictive control ; Probability theory ; Real time ; Reduced order models ; Regression analysis ; Smoothness ; Statistical analysis ; Tokamaks ; Uncertainty</subject><ispartof>Physics of plasmas, 2024-09, Vol.31 (9)</ispartof><rights>Author(s)</rights><rights>2024 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><cites>FETCH-LOGICAL-c209t-5dc463557c438e0c39a2bd8b24071939bf074f60899abf70110ebaa8e123d34b3</cites><orcidid>0000-0003-1937-2675 ; 0000-0002-2635-7118 ; 0000-0003-3742-1485 ; 0009-0001-8102-3962 ; 0000-0001-8817-4643 ; 0000-0001-9174-6227 ; 0000-0002-0292-5715 ; 0000-0003-4735-0991 ; 0000-0001-9520-8012 ; 0000-0002-8192-8411 ; 0000-0002-0437-8655 ; 0000000204378655 ; 0000000191746227 ; 0000000202925715 ; 0000000281928411 ; 0000000226357118 ; 0000000347350991 ; 0000000337421485 ; 0000000188174643 ; 0000000195208012 ; 0000000319372675 ; 0009000181023962</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>230,315,782,786,887,27933,27934</link.rule.ids><backlink>$$Uhttps://www.osti.gov/servlets/purl/2477694$$D View this record in Osti.gov$$Hfree_for_read</backlink></links><search><creatorcontrib>Madireddy, S.</creatorcontrib><creatorcontrib>Akçay, C.</creatorcontrib><creatorcontrib>Kruger, S. E.</creatorcontrib><creatorcontrib>Amara, T. Bechtel</creatorcontrib><creatorcontrib>Sun, X.</creatorcontrib><creatorcontrib>McClenaghan, J.</creatorcontrib><creatorcontrib>Koo, J.</creatorcontrib><creatorcontrib>Samaddar, A.</creatorcontrib><creatorcontrib>Liu, Y.</creatorcontrib><creatorcontrib>Balaprakash, P.</creatorcontrib><creatorcontrib>Lao, L. L.</creatorcontrib><creatorcontrib>General Atomics, San Diego, CA (United States)</creatorcontrib><title>EFIT-Prime: Probabilistic and physics-constrained reduced-order neural network model for equilibrium reconstruction in DIII-D</title><title>Physics of plasmas</title><description>We introduce EFIT-Prime, a novel machine learning surrogate model for EFIT (Equilibrium FIT) that integrates probabilistic and physics-informed methodologies to overcome typical limitations associated with deterministic and ad hoc neural network architectures. EFIT-Prime utilizes a neural architecture search-based deep ensemble for robust uncertainty quantification, providing scalable and efficient neural architectures that comprehensively quantify both data and model uncertainties. Physically informed by the Grad–Shafranov equation, EFIT-Prime applies a constraint on the current density Jtor and a smoothness constraint on the first derivative of the poloidal flux, ensuring physically plausible solutions. Furthermore, the spatial location of the diagnostics is explicitly incorporated in the inputs to account for their spatial correlation. Extensive evaluations demonstrate EFIT-Prime's accuracy and robustness across diverse scenarios, most notably showing good generalization on negative-triangularity discharges that were excluded from training. Timing studies indicate an ensemble inference time of 15 ms for predicting a new equilibrium, offering the possibility of plasma control in real-time, if the model is optimized for speed.</description><subject>Algorithms and data structure</subject><subject>Artificial neural networks</subject><subject>Bayesian statistics</subject><subject>Constraints</subject><subject>Equilibrium</subject><subject>Fusion energy</subject><subject>Machine learning</subject><subject>Magnetohydrodynamics</subject><subject>Neural networks</subject><subject>Nuclear fusion</subject><subject>Plasma control</subject><subject>Poloidal flux</subject><subject>Predictive control</subject><subject>Probability theory</subject><subject>Real time</subject><subject>Reduced order models</subject><subject>Regression analysis</subject><subject>Smoothness</subject><subject>Statistical analysis</subject><subject>Tokamaks</subject><subject>Uncertainty</subject><issn>1070-664X</issn><issn>1089-7674</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><recordid>eNp9kUtLxTAQhYso-Fz4D4KuFKKTJk0ad-KzIOhCwV3Iqxi9t7kmLeLC_24ude1qZvHNmTlzquqQwBkBTs-bM6gJ5SA3qh0CrcSCC7a57gVgztnrdrWb8zsAMN60O9XPzW33jJ9SWPoL9JSi0SYsQh6DRXpwaPX2nYPN2MYhj0mHwTuUvJusdzgm5xMa_JT0opTxK6YPtIzOL1AfE_KfU1EyKUzLMjILTHYMcUBhQNdd1-Hr_Wqr14vsD_7qXvVye_N8dY8fHu-6q8sHbGuQI26cZZw2jbCMth4slbo2rjU1A0EklaYHwXpe7EptegGEgDdat57U1FFm6F51NOvG4kxlG0Zv38pJg7ejqpkQXLICHc_QKsXPyedRvccpDeUuRQkQymQr2kKdzJRNMefke7Uqz9PpWxFQ6whUo_4iKOzpzK436rX1f-BftuuGJQ</recordid><startdate>20240901</startdate><enddate>20240901</enddate><creator>Madireddy, S.</creator><creator>Akçay, C.</creator><creator>Kruger, S. E.</creator><creator>Amara, T. Bechtel</creator><creator>Sun, X.</creator><creator>McClenaghan, J.</creator><creator>Koo, J.</creator><creator>Samaddar, A.</creator><creator>Liu, Y.</creator><creator>Balaprakash, P.</creator><creator>Lao, L. L.</creator><general>American Institute of Physics</general><general>American Institute of Physics (AIP)</general><scope>AJDQP</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>8FD</scope><scope>H8D</scope><scope>L7M</scope><scope>OIOZB</scope><scope>OTOTI</scope><orcidid>https://orcid.org/0000-0003-1937-2675</orcidid><orcidid>https://orcid.org/0000-0002-2635-7118</orcidid><orcidid>https://orcid.org/0000-0003-3742-1485</orcidid><orcidid>https://orcid.org/0009-0001-8102-3962</orcidid><orcidid>https://orcid.org/0000-0001-8817-4643</orcidid><orcidid>https://orcid.org/0000-0001-9174-6227</orcidid><orcidid>https://orcid.org/0000-0002-0292-5715</orcidid><orcidid>https://orcid.org/0000-0003-4735-0991</orcidid><orcidid>https://orcid.org/0000-0001-9520-8012</orcidid><orcidid>https://orcid.org/0000-0002-8192-8411</orcidid><orcidid>https://orcid.org/0000-0002-0437-8655</orcidid><orcidid>https://orcid.org/0000000204378655</orcidid><orcidid>https://orcid.org/0000000191746227</orcidid><orcidid>https://orcid.org/0000000202925715</orcidid><orcidid>https://orcid.org/0000000281928411</orcidid><orcidid>https://orcid.org/0000000226357118</orcidid><orcidid>https://orcid.org/0000000347350991</orcidid><orcidid>https://orcid.org/0000000337421485</orcidid><orcidid>https://orcid.org/0000000188174643</orcidid><orcidid>https://orcid.org/0000000195208012</orcidid><orcidid>https://orcid.org/0000000319372675</orcidid><orcidid>https://orcid.org/0009000181023962</orcidid></search><sort><creationdate>20240901</creationdate><title>EFIT-Prime: Probabilistic and physics-constrained reduced-order neural network model for equilibrium reconstruction in DIII-D</title><author>Madireddy, S. ; Akçay, C. ; Kruger, S. E. ; Amara, T. Bechtel ; Sun, X. ; McClenaghan, J. ; Koo, J. ; Samaddar, A. ; Liu, Y. ; Balaprakash, P. ; Lao, L. L.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c209t-5dc463557c438e0c39a2bd8b24071939bf074f60899abf70110ebaa8e123d34b3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Algorithms and data structure</topic><topic>Artificial neural networks</topic><topic>Bayesian statistics</topic><topic>Constraints</topic><topic>Equilibrium</topic><topic>Fusion energy</topic><topic>Machine learning</topic><topic>Magnetohydrodynamics</topic><topic>Neural networks</topic><topic>Nuclear fusion</topic><topic>Plasma control</topic><topic>Poloidal flux</topic><topic>Predictive control</topic><topic>Probability theory</topic><topic>Real time</topic><topic>Reduced order models</topic><topic>Regression analysis</topic><topic>Smoothness</topic><topic>Statistical analysis</topic><topic>Tokamaks</topic><topic>Uncertainty</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Madireddy, S.</creatorcontrib><creatorcontrib>Akçay, C.</creatorcontrib><creatorcontrib>Kruger, S. E.</creatorcontrib><creatorcontrib>Amara, T. Bechtel</creatorcontrib><creatorcontrib>Sun, X.</creatorcontrib><creatorcontrib>McClenaghan, J.</creatorcontrib><creatorcontrib>Koo, J.</creatorcontrib><creatorcontrib>Samaddar, A.</creatorcontrib><creatorcontrib>Liu, Y.</creatorcontrib><creatorcontrib>Balaprakash, P.</creatorcontrib><creatorcontrib>Lao, L. L.</creatorcontrib><creatorcontrib>General Atomics, San Diego, CA (United States)</creatorcontrib><collection>AIP Open Access Journals</collection><collection>CrossRef</collection><collection>Technology Research Database</collection><collection>Aerospace Database</collection><collection>Advanced Technologies Database with Aerospace</collection><collection>OSTI.GOV - Hybrid</collection><collection>OSTI.GOV</collection><jtitle>Physics of plasmas</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Madireddy, S.</au><au>Akçay, C.</au><au>Kruger, S. E.</au><au>Amara, T. Bechtel</au><au>Sun, X.</au><au>McClenaghan, J.</au><au>Koo, J.</au><au>Samaddar, A.</au><au>Liu, Y.</au><au>Balaprakash, P.</au><au>Lao, L. L.</au><aucorp>General Atomics, San Diego, CA (United States)</aucorp><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>EFIT-Prime: Probabilistic and physics-constrained reduced-order neural network model for equilibrium reconstruction in DIII-D</atitle><jtitle>Physics of plasmas</jtitle><date>2024-09-01</date><risdate>2024</risdate><volume>31</volume><issue>9</issue><issn>1070-664X</issn><eissn>1089-7674</eissn><coden>PHPAEN</coden><abstract>We introduce EFIT-Prime, a novel machine learning surrogate model for EFIT (Equilibrium FIT) that integrates probabilistic and physics-informed methodologies to overcome typical limitations associated with deterministic and ad hoc neural network architectures. EFIT-Prime utilizes a neural architecture search-based deep ensemble for robust uncertainty quantification, providing scalable and efficient neural architectures that comprehensively quantify both data and model uncertainties. Physically informed by the Grad–Shafranov equation, EFIT-Prime applies a constraint on the current density Jtor and a smoothness constraint on the first derivative of the poloidal flux, ensuring physically plausible solutions. Furthermore, the spatial location of the diagnostics is explicitly incorporated in the inputs to account for their spatial correlation. Extensive evaluations demonstrate EFIT-Prime's accuracy and robustness across diverse scenarios, most notably showing good generalization on negative-triangularity discharges that were excluded from training. Timing studies indicate an ensemble inference time of 15 ms for predicting a new equilibrium, offering the possibility of plasma control in real-time, if the model is optimized for speed.</abstract><cop>Melville</cop><pub>American Institute of Physics</pub><doi>10.1063/5.0213609</doi><tpages>20</tpages><orcidid>https://orcid.org/0000-0003-1937-2675</orcidid><orcidid>https://orcid.org/0000-0002-2635-7118</orcidid><orcidid>https://orcid.org/0000-0003-3742-1485</orcidid><orcidid>https://orcid.org/0009-0001-8102-3962</orcidid><orcidid>https://orcid.org/0000-0001-8817-4643</orcidid><orcidid>https://orcid.org/0000-0001-9174-6227</orcidid><orcidid>https://orcid.org/0000-0002-0292-5715</orcidid><orcidid>https://orcid.org/0000-0003-4735-0991</orcidid><orcidid>https://orcid.org/0000-0001-9520-8012</orcidid><orcidid>https://orcid.org/0000-0002-8192-8411</orcidid><orcidid>https://orcid.org/0000-0002-0437-8655</orcidid><orcidid>https://orcid.org/0000000204378655</orcidid><orcidid>https://orcid.org/0000000191746227</orcidid><orcidid>https://orcid.org/0000000202925715</orcidid><orcidid>https://orcid.org/0000000281928411</orcidid><orcidid>https://orcid.org/0000000226357118</orcidid><orcidid>https://orcid.org/0000000347350991</orcidid><orcidid>https://orcid.org/0000000337421485</orcidid><orcidid>https://orcid.org/0000000188174643</orcidid><orcidid>https://orcid.org/0000000195208012</orcidid><orcidid>https://orcid.org/0000000319372675</orcidid><orcidid>https://orcid.org/0009000181023962</orcidid><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier ISSN: 1070-664X
ispartof Physics of plasmas, 2024-09, Vol.31 (9)
issn 1070-664X
1089-7674
language eng
recordid cdi_osti_scitechconnect_2477694
source Alma/SFX Local Collection
subjects Algorithms and data structure
Artificial neural networks
Bayesian statistics
Constraints
Equilibrium
Fusion energy
Machine learning
Magnetohydrodynamics
Neural networks
Nuclear fusion
Plasma control
Poloidal flux
Predictive control
Probability theory
Real time
Reduced order models
Regression analysis
Smoothness
Statistical analysis
Tokamaks
Uncertainty
title EFIT-Prime: Probabilistic and physics-constrained reduced-order neural network model for equilibrium reconstruction in DIII-D
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-11-29T03%3A57%3A45IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_osti_&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=EFIT-Prime:%20Probabilistic%20and%20physics-constrained%20reduced-order%20neural%20network%20model%20for%20equilibrium%20reconstruction%20in%20DIII-D&rft.jtitle=Physics%20of%20plasmas&rft.au=Madireddy,%20S.&rft.aucorp=General%20Atomics,%20San%20Diego,%20CA%20(United%20States)&rft.date=2024-09-01&rft.volume=31&rft.issue=9&rft.issn=1070-664X&rft.eissn=1089-7674&rft.coden=PHPAEN&rft_id=info:doi/10.1063/5.0213609&rft_dat=%3Cproquest_osti_%3E3101349878%3C/proquest_osti_%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=3101349878&rft_id=info:pmid/&rfr_iscdi=true