Hao Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hao Wu has a total of 79 co-authors. Here are the top co-authors by number of shared publications:
- Didier Mazon
CEA, IRFM, F-13108 Saint Paul Lez Durance, France
3 shared publications - Geert Verdoolaege
Ghent University
3 shared publications - Axel Jardin
Institute of Nuclear Physics Polish Academy of Sciences (IFJ PAN), PL-31-342 Krakow, Poland
2 shared publications A Jardin
Institute of Nuclear Physics Polish Academy of Sciences (IFJ PAN), 31-342 Kraków, Poland
1 shared publicationA Wojenski
Warsaw University of Technology, Institute of Electronic Systems, Nowowiejska 15/19, 00-665 Warsaw, Poland
1 shared publicationA. Puig Sitjes
Max Planck Institut fuer Plasmaphysik, Garching, Germany
1 shared publication
Latest Publications
- Open Access
- 22 October 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 67 (1) , 016004
Real-time CXRS analysis on HL-3 tokamak with reliable neural networks and cross-device transfer learning
We present a neural network-based method for real-time charge exchange recombination spectroscopy (CXRS) analysis on the HL-3 tokamak. To enhance model performance and reliability, we tailor the network architecture to the physical characteristics of CXRS spectra: the receptive field matches the Doppler-broadened line width, channel embedding encodes spatial awareness across diagnostic channels, and attention pooling focuses on the most informative spectral regions. These designs yield millisecond-level inference (0.56 ms per 32-channel) with accuracy comparable to conventional methods ( R 2 _ T i = 0.93, R 2 _ v t = 0.98). Furthermore, the model’s reliability is established through interpretability and uncertainty analysis. Interpretability analysis confirms that the model focuses on the Doppler-broadened wings of C VI for T i and tracks line shifts for v t , consistent with physical principles. Uncertainty estimates show that the model appropriately expresses higher uncertainty in regions where predictions are inherently more challenging. To address data scarcity on new devices, transfer learning from HL-2A tokamak markedly boosts performance in low-data regimes while retaining a modest advantage even with full HL-3 data, provided that the spectra are properly aligned and baseline subtracted. This work establishes a real-time, reliable, and data-efficient AI framework for CXRS analysis, offering a practical path to deploy machine learning approaches in plasma control on future fusion devices.
- 22 October 2026
- Published by IOP Publishing in Nuclear Fusion
- Vol. 67 (1) , 016004
Open AccessReal-time CXRS analysis on HL-3 tokamak with reliable neural networks and cross-device transfer learning
We present a neural network-based method for real-time charge exchange recombination spectroscopy (CXRS) analysis on the HL-3 tokamak. To enhance model performance and reliability, we tailor the network architecture to the physical characteristics of CXRS spectra: the receptive field matches the Doppler-broadened line width, channel embedding encodes spatial awareness across diagnostic channels, and attention pooling focuses on the most informative spectral regions. These designs yield millisecond-level inference (0.56 ms per 32-channel) with accuracy comparable to conventional methods ( R 2 _ T i = 0.93, R 2 _ v t = 0.98). Furthermore, the model’s reliability is established through interpretability and uncertainty analysis. Interpretability analysis confirms that the model focuses on the Doppler-broadened wings of C VI for T i and tracks line shifts for v t , consistent with physical principles. Uncertainty estimates show that the model appropriately expresses higher uncertainty in regions where predictions are inherently more challenging. To address data scarcity on new devices, transfer learning from HL-2A tokamak markedly boosts performance in low-data regimes while retaining a modest advantage even with full HL-3 data, provided that the spectra are properly aligned and baseline subtracted. This work establishes a real-time, reliable, and data-efficient AI framework for CXRS analysis, offering a practical path to deploy machine learning approaches in plasma control on future fusion devices.
- Open Access
- 27 November 2025
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (1) , 017002
Summary of the 5th IAEA technical meeting on fusion data processing, validation and analysis (FDPVA)
The purpose of the 5th International Atomic Energy Agency technical meeting on fusion data processing, validation and analysis (FDPVA) (Ghent University, Ghent, Belgium, 12–15 June 2023) was to provide a platform during which a set of topics relevant to FDPVA were discussed with the view of meeting the needs of next step fusion devices such as ITER. The validation and analysis of experimental data obtained from diagnostics used to characterize fusion plasmas are crucial for a knowledge-based understanding of the physical processes governing the dynamics of these plasmas. This paper presents the recent progress and achievements in the domain of plasma diagnostics data analysis and synthetic diagnostics reported at the meeting, including concept description of new devices; fusion databases; integrated data analysis; inverse problems; uncertainty propagation, verification and validation; probabilistic methods and machine learning. The relevant results underline trends observed in the current major fusion confinement devices.
- 27 November 2025
- Published by IOP Publishing in Nuclear Fusion
- Vol. 66 (1) , 017002
Open AccessSummary of the 5th IAEA technical meeting on fusion data processing, validation and analysis (FDPVA)
The purpose of the 5th International Atomic Energy Agency technical meeting on fusion data processing, validation and analysis (FDPVA) (Ghent University, Ghent, Belgium, 12–15 June 2023) was to provide a platform during which a set of topics relevant to FDPVA were discussed with the view of meeting the needs of next step fusion devices such as ITER. The validation and analysis of experimental data obtained from diagnostics used to characterize fusion plasmas are crucial for a knowledge-based understanding of the physical processes governing the dynamics of these plasmas. This paper presents the recent progress and achievements in the domain of plasma diagnostics data analysis and synthetic diagnostics reported at the meeting, including concept description of new devices; fusion databases; integrated data analysis; inverse problems; uncertainty propagation, verification and validation; probabilistic methods and machine learning. The relevant results underline trends observed in the current major fusion confinement devices.
- 25 July 2025
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 67 (8) , 085001
Bayesian integrated estimation of two-dimensional tungsten concentration profiles at WEST using soft x-ray and bolometry diagnostics
Tungsten accumulation in the core of tokamak plasmas poses a major risk to the core performance through strong radiative cooling. Reliable estimates of tungsten concentrations are therefore critical towards fusion-relevant operation. In this article, we present Bayesian integrated data analysis (IDA) on measurements from soft x-ray, interferometry and electron cyclotron emission diagnostics at WEST, to jointly estimate the two-dimensional tungsten concentration profiles and kinetic profiles ( n e and T e ). IDA aims at providing reliable parameter estimates that are consistent with all available measurements, together with adequate uncertainty quantification. The IDA approach relies on the exploration of a high-dimensional joint posterior distribution of the quantities of interest, which, in our work, is implemented efficiently via Markov chain Monte Carlo sampling techniques with reparameterization. This method has been tested on both synthetic and experimental data. It yields reliable tungsten concentration estimates in the core plasma, whereas the uncertainty strongly increases towards the edge. Therefore, we also incorporate bolometry measurements, with first results indicating the strong potential of this method.
- 25 July 2025
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 67 (8) , 085001
Bayesian integrated estimation of two-dimensional tungsten concentration profiles at WEST using soft x-ray and bolometry diagnostics
Tungsten accumulation in the core of tokamak plasmas poses a major risk to the core performance through strong radiative cooling. Reliable estimates of tungsten concentrations are therefore critical towards fusion-relevant operation. In this article, we present Bayesian integrated data analysis (IDA) on measurements from soft x-ray, interferometry and electron cyclotron emission diagnostics at WEST, to jointly estimate the two-dimensional tungsten concentration profiles and kinetic profiles ( n e and T e ). IDA aims at providing reliable parameter estimates that are consistent with all available measurements, together with adequate uncertainty quantification. The IDA approach relies on the exploration of a high-dimensional joint posterior distribution of the quantities of interest, which, in our work, is implemented efficiently via Markov chain Monte Carlo sampling techniques with reparameterization. This method has been tested on both synthetic and experimental data. It yields reliable tungsten concentration estimates in the core plasma, whereas the uncertainty strongly increases towards the edge. Therefore, we also incorporate bolometry measurements, with first results indicating the strong potential of this method.
- Open Access
- 26 June 2025
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 67 (7) , 075012
Reconstruction of tungsten concentration in WEST plasma core with Machine Learning
In modern tokamaks like ITER or WEST, with tungsten (W) instead of traditional carbon as the main plasma-facing material to minimize erosion and tritium retention in the walls, the essential issue of heavy impurity radiation has been raised. Monitoring and real-time control of W concentration below 0.01% in the plasma core will be indeed necessary to avoid significant cooling of the plasma by impurity radiation, in particular in the soft x-ray (SXR) energy range of 0.1–20 keV, and to select adequate mitigation strategies. In this context, this paper describes recent work to reconstruct W concentration in WEST plasma core from SXR, electron profiles measurements and magnetic equilibrium in a fast and automatized way, with the support of machine learning. A significant data reduction is performed, with the parametrization of the magnetic equilibrium and electron radial profiles, to limit the number of inputs feeding the considered neural network (NN). The NN architecture, training strategy and experimental dataset used to perform fast reconstructions are introduced. The NN is trained with 2023 WEST data and tested on a set of 2024 plasma discharges. NN predictions are compared with the ones calculated with a synthetic diagnostic tool for different ranges of plasma temperature and density. Perspectives for further improvements and extension of this study are also discussed.
- 26 June 2025
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 67 (7) , 075012
Open AccessReconstruction of tungsten concentration in WEST plasma core with Machine Learning
In modern tokamaks like ITER or WEST, with tungsten (W) instead of traditional carbon as the main plasma-facing material to minimize erosion and tritium retention in the walls, the essential issue of heavy impurity radiation has been raised. Monitoring and real-time control of W concentration below 0.01% in the plasma core will be indeed necessary to avoid significant cooling of the plasma by impurity radiation, in particular in the soft x-ray (SXR) energy range of 0.1–20 keV, and to select adequate mitigation strategies. In this context, this paper describes recent work to reconstruct W concentration in WEST plasma core from SXR, electron profiles measurements and magnetic equilibrium in a fast and automatized way, with the support of machine learning. A significant data reduction is performed, with the parametrization of the magnetic equilibrium and electron radial profiles, to limit the number of inputs feeding the considered neural network (NN). The NN architecture, training strategy and experimental dataset used to perform fast reconstructions are introduced. The NN is trained with 2023 WEST data and tested on a set of 2024 plasma discharges. NN predictions are compared with the ones calculated with a synthetic diagnostic tool for different ranges of plasma temperature and density. Perspectives for further improvements and extension of this study are also discussed.