Density limit disruption prediction using a long short-term memory network on EAST

Zhang, Kai; Chen, Dalong; Guo, Bihao; Chen, Junjie; Xiao, Bingjia

doi:10.1088/2058-6272/abb28f

Cited by 6 publications

(5 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Additionally, neural networks are instrumental in establishing predictive models to forecast the reaction process and outcomes of fusion energy. For example, long short-term memory networks can be utilized to develop predictive models for forecasting the energy output and reaction speed of fusion energy [19][20][21][22][23][24][25]. Moreover, neural networks play a pivotal role in anomaly detection and fault diagnosis within fusion energy devices.…”

Section: Neural Network and Nuclear Fusionmentioning

confidence: 99%

Static performance prediction of long-pulse negative ion based neutral beam injection experiment

Li,

Hu,

Zhao

et al. 2024

Plasma Phys. Control. Fusion

View full text Add to dashboard Cite

The mission of Negative Ion-Based Neutral Beam Injection (NNBI) is to conduct experiments with pulses lasting thousands of seconds. It is crucial to develop a simplified physical calculation model for the long-pulse negative ion source in the current NNBI device. This model will be used to evaluate the advantages and disadvantages of the selected parameters prior to the experiment, and to assist in adjusting and establishing the experimental parameters for the long-pulse ion source experiment. This paper presents the development of a static performance prediction model using a Back Propagation (BP) neural network. The model assesses the yield of negative hydrogen ions and the quantity of electrons in the ion source under specific parameter conditions, utilizing various experimental parameters as input. The experimental data used for this model are derived from historical data generated during the operation of the 2022 NNBI experiment. The test results indicate that under the current optimal hyperparameter condition, the prediction accuracy of I_H- is 80.84%, and the prediction accuracy of I_EG is 77.57%. This can effectively prevent invalid shots, accurately assess the advantages and disadvantages of the input parameters, and enhance the performance of the long-pulse NNBI device.

show abstract

Section: Neural Network and Nuclear Fusionmentioning

confidence: 99%

Static performance prediction of long-pulse negative ion based neutral beam injection experiment

Li,

Hu,

Zhao

et al. 2024

Plasma Phys. Control. Fusion

View full text Add to dashboard Cite

show abstract

“…However, it took a very long time to train the model for the entire flat-top phase. To resolve this problem, the LSTM approach [47] based on the short time sequence has been implemented for an EAST tokamak. This approach helps to minimize the computational time as well as to fade the gradient problem that occurred in the model when it was trained by the RNN.…”

Section: Sliding Window Mechanismmentioning

confidence: 99%

A Review of Traditional and Data-Driven Approaches for Disruption Prediction in Different Tokamaks

Priyanka,

Sangeetha,

Jayakumar

2024

E3S Web Conf.

View full text Add to dashboard Cite

Tokamak is a nuclear fusion reactor; inside, the two lighter nuclei known as deuterium and tritium are first ionized together to form plasma, which is heated up to 150 million degrees Celsius, and then they are confined by the torus-shaped magnetic field. During this process, it releases a massive amount of energy, making fusion a feasible option for a long-term and renewable source of energy. On the other hand, plasma leads to disruptions as a consequence of the sudden implosion of the system, which halts the fusion process. Disruptions can irrevocably harm current fusion devices and are predicted to have a more catastrophic impact on feature devices such as ITER since they cause a rapid loss of confinement. To control, and prevent disruptions, or at least lessen their negative impact by mitigating them, various traditional and data-driven models obtained with machine learning and deep learning techniques have been used, an overview of some of which is presented in this article. These models are commonly used to forecast their occurrence and give sufficient time to take some counteractive measures.

show abstract

“…ML techniques have begun to be applied to the density limit problem as well [12], for example using a radiation profile measurement as input for a neural network [13] or decision tree [14] in J-TEXT or using a long short-term memory network [15] or a random forest [16] in EAST. Additionally, Greenwald fraction (n e /n G ) has been used as an input to random forest based [17] or deep learning [18] disruption prediction algorithms run on multiple machines, and it was found that it can be an important contributor [19].…”

Section: Introductionmentioning

confidence: 99%

Density limits as disruption forecasters for spherical tokamaks

Berkery

Sabbagh

Ham

et al. 2023

Plasma Phys. Control. Fusion

View full text Add to dashboard Cite

Fusion power output from spherical tokamaks would benefit from increased confined plasma density, but there exists a limit on the density before confinement is lost and the plasma current is disrupted. This density limit has long been characterized by a simple, global Greenwald limit proportional to the plasma current and inversely proportional to the cross sectional area of the plasma. It is shown that in the database of discharges from the NSTX and MAST spherical tokamaks, the likelihood of disruption does increase above the Greenwald limit, and especially in the plasma current rampdown phase. The physics of the density limit has been recently theoretically explored through local criteria. Several of these are tested using the disruption event characterization and forecasting (DECAFTM) code for their potential effectiveness as disruption warning signals. For a limited set of NSTX discharges, a local island power balance criteria was found to be less reliable, presently, than the Greenwald limit. An empirical critical edge line density and a boundary turbulent transport limit were both tested for MAST-U, which has an electron density profile measurement with high spatial resolution in the outer part of the plasma. Both were found to have similar dependencies on key plasma parameters. In a limited set of MAST-U discharges that appear to disrupt due to rising density at values under the Greenwald limit, crossing of the boundary turbulent transport limit occurred close to the time of disruption. Finally, these limits were evaluated for their potential use in real-time, and it was found that with the necessary real-time inputs and with refinement through further testing, these limits could be implemented in a real-time disruption forecasting system.

show abstract

Density limit disruption prediction using a long short-term memory network on EAST

Cited by 6 publications

References 39 publications

Static performance prediction of long-pulse negative ion based neutral beam injection experiment

Static performance prediction of long-pulse negative ion based neutral beam injection experiment

A Review of Traditional and Data-Driven Approaches for Disruption Prediction in Different Tokamaks

Density limits as disruption forecasters for spherical tokamaks

Contact Info

Product

Resources

About