Zixian Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Zixian Wu has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Honglei Jin
School of Computer Science Jiangsu University of Science and Technology Zhenjiang China
5 shared publications - Hongwei Ye
School of Computer Science Jiangsu University of Science and Technology Zhenjiang China
5 shared publications - Noraisyah Mohamed Shah
Department Of Electrical Engineering, Faculty of Engineering, Universiti Malaya, 50603 Kuala Lumpur, Malaysia
5 shared publications - Shunhuang Zhang
School of Computer Science Jiangsu University of Science and Technology Zhenjiang China
5 shared publications - Xuebao Li
School of Computer Science Jiangsu University of Science and Technology Zhenjiang China
5 shared publications - Xusheng Huang
School of Computer Science Jiangsu University of Science and Technology Zhenjiang China
5 shared publications
Latest Publications
- Open Access
- 14 October 2026
- Published by American Geophysical Union (AGU) in Space Weather
- Vol. 24 (10)
A Near‐Real‐Time Deep‐Learning Solar Flare Forecasting System Supported by SDO/HMI and ASO‐S/FMG Magnetograms
This study develops and evaluates five deep learning models, including image‐based (CNN, SwinT) and video‐based (CNN‐LSTM, SwinT‐LSTM, Video SwinT), to forecast M‐class flares within the next 24 hr using cross‐validation (CV) data sets. Based on these models, we develop a near‐real‐time (NRT) flare forecasting system that issues daily predictions at 00:00 UTC. We compare model performance during development and in the NRT system to assess temporal generalization. We then compare daily forecasting performance of our system with that of NASA/CCMC and SolarFlareNet. Moreover, ASO‐S/FMG magnetograms are ingested daily into our system, with their predictions compared with SHARP‐based predictions to assess generalization across data products. Key results include: (a) On the CV data sets, SwinT‐LSTM achieves the highest mean TSS and BSS, comparable to SwinT and Video SwinT. (b) SwinT‐based models outperform CNN‐based models, and video‐based models generally achieve higher mean scores than image‐based models but are statistically comparable. (c) Video SwinT achieves competitive performance in our NRT system, demonstrating strong temporal generalization in categorical forecasting. In matched comparisons with NASA/CCMC and SolarFlareNet, our system shows competitive performance in daily M‐class flare forecasting within of the central meridian. (d) The SHARP‐based system generalizes well to ASO‐S/FMG, with SwinT showing the strongest generalization ability in categorical forecasting among the five models. To our knowledge, we are the first to introduce SwinT‐LSTM and Video SwinT into flare forecasting, and their deployment in our system shows improvements over existing platforms. Moreover, we are the first to incorporate ASO‐S/FMG magnetograms into an NRT flare forecasting system.
- 14 October 2026
- Published by American Geophysical Union (AGU) in Space Weather
- Vol. 24 (10)
Open AccessA Near‐Real‐Time Deep‐Learning Solar Flare Forecasting System Supported by SDO/HMI and ASO‐S/FMG Magnetograms
This study develops and evaluates five deep learning models, including image‐based (CNN, SwinT) and video‐based (CNN‐LSTM, SwinT‐LSTM, Video SwinT), to forecast M‐class flares within the next 24 hr using cross‐validation (CV) data sets. Based on these models, we develop a near‐real‐time (NRT) flare forecasting system that issues daily predictions at 00:00 UTC. We compare model performance during development and in the NRT system to assess temporal generalization. We then compare daily forecasting performance of our system with that of NASA/CCMC and SolarFlareNet. Moreover, ASO‐S/FMG magnetograms are ingested daily into our system, with their predictions compared with SHARP‐based predictions to assess generalization across data products. Key results include: (a) On the CV data sets, SwinT‐LSTM achieves the highest mean TSS and BSS, comparable to SwinT and Video SwinT. (b) SwinT‐based models outperform CNN‐based models, and video‐based models generally achieve higher mean scores than image‐based models but are statistically comparable. (c) Video SwinT achieves competitive performance in our NRT system, demonstrating strong temporal generalization in categorical forecasting. In matched comparisons with NASA/CCMC and SolarFlareNet, our system shows competitive performance in daily M‐class flare forecasting within of the central meridian. (d) The SHARP‐based system generalizes well to ASO‐S/FMG, with SwinT showing the strongest generalization ability in categorical forecasting among the five models. To our knowledge, we are the first to introduce SwinT‐LSTM and Video SwinT into flare forecasting, and their deployment in our system shows improvements over existing platforms. Moreover, we are the first to incorporate ASO‐S/FMG magnetograms into an NRT flare forecasting system.
- Open Access
- 27 March 2026
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 283 (2) , 71
F10.7 Index Prediction: A Multiscale Decomposition Strategy with Wavelet Transform for Performance Optimization
In this study, we construct Dataset A for training, validation, and testing, and Dataset B to evaluate generalization. We propose a novel F10.7 index forecasting method using wavelet decomposition, which feeds F10.7 together with its decomposed approximate and detail signals into the iTransformer model. We also incorporate the International Sunspot Number (ISN) and its wavelet-decomposed signals to assess their influence on prediction performance. Our optimal method is then compared with the latest method from S. Yan et al. and three operational models (Space Weather Prediction Center, British Geological Survey, Collect Localization Satellites). Additionally, we transfer our method to the PatchTST model used in H. Ye et al. and compare our method with theirs on Dataset B. Key findings include the following: (1) The wavelet-based combination methods overall outperform the baseline using only the F10.7 index. The prediction performance improves as higher-level approximate and detail signals are incrementally added. The Combination 6 method—integrating F10.7 with its first to fifth level approximate and detail signals—outperforms methods using only approximate or detail signals. (2) Incorporating the ISN and its wavelet-decomposed signals does not enhance prediction performance. (3) The Combination 6 method significantly surpasses that of S. Yan et al. and three operational models, with rms error, mean absolute error, and mean absolute percentage error reduced by 18.22%, 15.09%, and 8.57%, respectively, against the former method. It also excels across four different conditions of solar activity. (4) Our method demonstrates superior generalization and prediction capability over the method of H. Ye et al. across all forecast horizons. To our knowledge, this is the first application of wavelet decomposition in F10.7 prediction, substantially improving forecast performance.
- 27 March 2026
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 283 (2) , 71
Open AccessF10.7 Index Prediction: A Multiscale Decomposition Strategy with Wavelet Transform for Performance Optimization
In this study, we construct Dataset A for training, validation, and testing, and Dataset B to evaluate generalization. We propose a novel F10.7 index forecasting method using wavelet decomposition, which feeds F10.7 together with its decomposed approximate and detail signals into the iTransformer model. We also incorporate the International Sunspot Number (ISN) and its wavelet-decomposed signals to assess their influence on prediction performance. Our optimal method is then compared with the latest method from S. Yan et al. and three operational models (Space Weather Prediction Center, British Geological Survey, Collect Localization Satellites). Additionally, we transfer our method to the PatchTST model used in H. Ye et al. and compare our method with theirs on Dataset B. Key findings include the following: (1) The wavelet-based combination methods overall outperform the baseline using only the F10.7 index. The prediction performance improves as higher-level approximate and detail signals are incrementally added. The Combination 6 method—integrating F10.7 with its first to fifth level approximate and detail signals—outperforms methods using only approximate or detail signals. (2) Incorporating the ISN and its wavelet-decomposed signals does not enhance prediction performance. (3) The Combination 6 method significantly surpasses that of S. Yan et al. and three operational models, with rms error, mean absolute error, and mean absolute percentage error reduced by 18.22%, 15.09%, and 8.57%, respectively, against the former method. It also excels across four different conditions of solar activity. (4) Our method demonstrates superior generalization and prediction capability over the method of H. Ye et al. across all forecast horizons. To our knowledge, this is the first application of wavelet decomposition in F10.7 prediction, substantially improving forecast performance.
- Open Access
- 19 February 2026
- Published by American Geophysical Union (AGU) in Space Weather
- Vol. 24 (2)
Operational Solar Flare Forecasting System Using an Explainable Large Language Model
This study focuses on forecasting major (M‐class) solar flares that can severely impact the near‐Earth environment. We construct two types of data sets using the Space Weather HMI Active Region Patches (SHARP), and develop a flare prediction network based on large language model (LLMFlareNet). We apply SHapley Additive exPlanations (SHAP) to explain the model predictions. We develop an operational forecasting system based on the LLMFlareNet model. We adopt a daily mode for performance comparison across various operational forecasting systems under identical active region (AR) number and prediction date, using daily operational observational data. The main results are as follows. (a) Through ablation experiments and comparison with baseline models, LLMFlareNet achieves the best TSS scores of 0.720 0.040 on the ten cross‐validation (CV) data set with mixed ARs. (b) By both global and local SHAP analyses, we identify that R_VALUE is the most influential physical feature for the prediction of LLMFlareNet, aligning with flare magnetic reconnection theory. (c) In daily mode, LLMFlareNet achieves TSS scores of 0.680/0.571 (0.689/0.661, respectively) on the data set with single/mixed ARs, markedly outperforming NASA/CCMC (SolarFlareNet, respectively). This work introduces the first application of a large language model as a universal computation engine with explainability method in this domain, and presents the first comparison between operational flare forecasting systems in daily mode. The proposed LLMFlareNet‐based system demonstrates substantial improvements over existing systems.
- 19 February 2026
- Published by American Geophysical Union (AGU) in Space Weather
- Vol. 24 (2)
Open AccessOperational Solar Flare Forecasting System Using an Explainable Large Language Model
This study focuses on forecasting major (M‐class) solar flares that can severely impact the near‐Earth environment. We construct two types of data sets using the Space Weather HMI Active Region Patches (SHARP), and develop a flare prediction network based on large language model (LLMFlareNet). We apply SHapley Additive exPlanations (SHAP) to explain the model predictions. We develop an operational forecasting system based on the LLMFlareNet model. We adopt a daily mode for performance comparison across various operational forecasting systems under identical active region (AR) number and prediction date, using daily operational observational data. The main results are as follows. (a) Through ablation experiments and comparison with baseline models, LLMFlareNet achieves the best TSS scores of 0.720 0.040 on the ten cross‐validation (CV) data set with mixed ARs. (b) By both global and local SHAP analyses, we identify that R_VALUE is the most influential physical feature for the prediction of LLMFlareNet, aligning with flare magnetic reconnection theory. (c) In daily mode, LLMFlareNet achieves TSS scores of 0.680/0.571 (0.689/0.661, respectively) on the data set with single/mixed ARs, markedly outperforming NASA/CCMC (SolarFlareNet, respectively). This work introduces the first application of a large language model as a universal computation engine with explainability method in this domain, and presents the first comparison between operational flare forecasting systems in daily mode. The proposed LLMFlareNet‐based system demonstrates substantial improvements over existing systems.
- Open Access
Operational Solar Flare Forecasting System Using an Explainable Large Language Model
This study focuses on forecasting major (>=M-class) solar flares that can severely impact the near-Earth environment. We construct two types of datasets using the Space Weather HMI Active Region Patches (SHARP), and develop a flare prediction network based on large language model (LLMFlareNet). We apply SHapley Additive exPlanations (SHAP) to explain the model predictions. We develop an operational forecasting system based on the LLMFlareNet model. We adopt a daily mode for performance comparison across various operational forecasting systems under identical active region (AR) number and prediction date, using daily operational observational data. The main results are as follows. (1) Through ablation experiments and comparison with baseline models, LLMFlareNet achieves the best TSS scores of 0.720 +/- 0.040 on the ten cross-validation (CV) dataset with mixed ARs. (2) By both global and local SHAP analyses, we identify that R_VALUE is the most influential physical feature for the prediction of LLMFlareNet, aligning with flare magnetic reconnection theory. (3) In daily mode, LLMFlareNet achieves TSS scores of 0.680/0.571 (0.689/0.661, respectively) on the dataset with single/mixed ARs, markedly outperforming NASA/CCMC (SolarFlareNet, respectively). This work introduces the first application of a large language model as a universal computation engine with explainability method in this domain, and presents the first comparison between operational flare forecasting systems in daily mode. The proposed LLMFlareNet-based system demonstrates substantial improvements over existing systems.
Open AccessOperational Solar Flare Forecasting System Using an Explainable Large Language Model
This study focuses on forecasting major (>=M-class) solar flares that can severely impact the near-Earth environment. We construct two types of datasets using the Space Weather HMI Active Region Patches (SHARP), and develop a flare prediction network based on large language model (LLMFlareNet). We apply SHapley Additive exPlanations (SHAP) to explain the model predictions. We develop an operational forecasting system based on the LLMFlareNet model. We adopt a daily mode for performance comparison across various operational forecasting systems under identical active region (AR) number and prediction date, using daily operational observational data. The main results are as follows. (1) Through ablation experiments and comparison with baseline models, LLMFlareNet achieves the best TSS scores of 0.720 +/- 0.040 on the ten cross-validation (CV) dataset with mixed ARs. (2) By both global and local SHAP analyses, we identify that R_VALUE is the most influential physical feature for the prediction of LLMFlareNet, aligning with flare magnetic reconnection theory. (3) In daily mode, LLMFlareNet achieves TSS scores of 0.680/0.571 (0.689/0.661, respectively) on the dataset with single/mixed ARs, markedly outperforming NASA/CCMC (SolarFlareNet, respectively). This work introduces the first application of a large language model as a universal computation engine with explainability method in this domain, and presents the first comparison between operational flare forecasting systems in daily mode. The proposed LLMFlareNet-based system demonstrates substantial improvements over existing systems.
- Open Access
- 28 January 2026
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 282 (2) , 52
A TimesNet-based Real-time Forecasting System for the F10.7 Index Using DRAO and Chinese Langfang Datasets
In this paper, we construct the DRAO univariate dataset, the DRAO multivariate dataset, and the Chinese Langfang dataset. We develop and compare five deep learning models (TimesNet, iTransformer, PatchTST, N -Beats, BiGRU) and a benchmark artificial neural network (ANN) model to predict the F10.7 index. We study the impact of different feature combinations on the performance of the recommended TimesNet model. Furthermore, we develop a real-time forecasting system for the F10.7 index, incorporating both univariate and multivariate TimesNet models. During the same period, we compare F10.7 prediction performance between our system and that of four foreign institutions (British Geological Survey (BGS), SWPC, Collecte Localisation Satellites (CLS), DRAO). We conduct daily averaged and hourly resolution forecasting using the Langfang dataset. To our knowledge, we establish the first TimesNet-based framework for F10.7 prediction, advancing hourly resolution F10.7 forecasting for the first time. The main results are as follows. (1) The univariate TimesNet model achieves superior prediction performance on the first to the 27th day of forecasting, outperforming both four deep learning models and the ANN model. With the increase in the prediction days, the prediction performance of the six models all shows a downward trend. (2) The multivariate TimesNet-FIAC model, using optimal feature combinations, outperforms the univariate TimesNet-F model. (3) In short-term prediction, TimesNet-FIAC within our system surpasses four foreign institutions. On the first forecasting day, its root mean square error, mean absolute error, and mean absolute percentage error decrease by 15.06%, 18.54%, and 20.90% compared to BGS, and by 3.54%, 10.21%, and 14.94% compared to CLS. (4) On the Langfang dataset, TimesNet-F demonstrates superior generalization in daily averaged short-term forecasting, and maintains good and stable performance in hourly resolution short-term prediction.
- 28 January 2026
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 282 (2) , 52
Open AccessA TimesNet-based Real-time Forecasting System for the F10.7 Index Using DRAO and Chinese Langfang Datasets
In this paper, we construct the DRAO univariate dataset, the DRAO multivariate dataset, and the Chinese Langfang dataset. We develop and compare five deep learning models (TimesNet, iTransformer, PatchTST, N -Beats, BiGRU) and a benchmark artificial neural network (ANN) model to predict the F10.7 index. We study the impact of different feature combinations on the performance of the recommended TimesNet model. Furthermore, we develop a real-time forecasting system for the F10.7 index, incorporating both univariate and multivariate TimesNet models. During the same period, we compare F10.7 prediction performance between our system and that of four foreign institutions (British Geological Survey (BGS), SWPC, Collecte Localisation Satellites (CLS), DRAO). We conduct daily averaged and hourly resolution forecasting using the Langfang dataset. To our knowledge, we establish the first TimesNet-based framework for F10.7 prediction, advancing hourly resolution F10.7 forecasting for the first time. The main results are as follows. (1) The univariate TimesNet model achieves superior prediction performance on the first to the 27th day of forecasting, outperforming both four deep learning models and the ANN model. With the increase in the prediction days, the prediction performance of the six models all shows a downward trend. (2) The multivariate TimesNet-FIAC model, using optimal feature combinations, outperforms the univariate TimesNet-F model. (3) In short-term prediction, TimesNet-FIAC within our system surpasses four foreign institutions. On the first forecasting day, its root mean square error, mean absolute error, and mean absolute percentage error decrease by 15.06%, 18.54%, and 20.90% compared to BGS, and by 3.54%, 10.21%, and 14.94% compared to CLS. (4) On the Langfang dataset, TimesNet-F demonstrates superior generalization in daily averaged short-term forecasting, and maintains good and stable performance in hourly resolution short-term prediction.