Search

[Introduction] Plasma disruption poses a significant threat to the tokamak nuclear device during its running and can cause damage to the device. Such damage can be reduced by adopting the disruption mitigation system, which has an action time highly dependent on the real-time running plasma disruption prediction system for predicting the plasma disruption moment. The deep-learning-based neural network has been used to train plasma disruption prediction models, and the real-time running of the deep-learning-based disruption prediction models requires a huge amount of real-time data from multiple diagnostics. [Method] The article proposed a design scheme for a real-time data acquisition system. The real-time data acquisition and transmission system was designed based on the modular structure and divided into the multiple channels acquisition module, ADC converting control and data reading module, data grouping and packing module and data transmission network module. The data transmission network module was developed on the hardware UDP network stack running on the FPGA at a speed of 10 G. This hardware UDP network stack featured a deterministic data transmission process, enabling a very low transmission latency of the system. [Result] The real-time data acquisition system has a sampling rate reaching 2 MSa/s, a data throughput rate exceeding 9.3 Gb/s, and a data transmission latency of less than 10 μs. [Conclusion] This data acquisition system facilitates the fast transmission of diagnostic data streams to disruption prediction models. The high sampling rates enable the system to perform real-time transmission of one-dimensional diagnostics such as radiation and electron temperature, improving the temporal resolution of data. The high data throughput rate can increase the transmission volume of diagnostic data, and the low data transmission latency can reduce the time required for disruption prediction models to obtain diagnostics data.

Plasma disruption in tokamak experiments is a challenging issue that causes damage to the device. Reliable prediction methods are needed, but the lack of full understanding of plasma disruption limits the effectiveness of physics-driven methods. Data-driven methods based on supervised learning are commonly used, and they rely on labelled training data. However, manual labelling of disruption precursors is a time-consuming and challenging task, as some precursors are difficult to accurately identify. The mainstream labelling methods assume that the precursor onset occurs at a fixed time before disruption, which leads to mislabeled samples and suboptimal prediction performance. In this paper, we present disruption prediction methods based on anomaly detection to address these issues, demonstrating good prediction performance on J-TEXT and EAST. By evaluating precursor onset times using different anomaly detection algorithms, it is found that labelling methods can be improved since the onset times of different shots are not necessarily the same. The study optimizes precursor labelling using the onset times inferred by the anomaly detection predictor and test the optimized labels on supervised learning disruption predictors. The results on J-TEXT and EAST show that the models trained on the optimized labels outperform those trained on fixed onset time labels.

Plasma disruption presents a significant challenge in tokamak fusion, especially in large-size devices like ITER, where it causes severe damage. While current data-driven machine learning methods perform well in disruption prediction, they require extensive discharge data for model training. However, future tokamaks will begin operations without any prior data, making it difficult to train data-driven disruption predictors and select appropriate hyperparameters during the early operation period. In this period disruption prediction also aims to support safe exploration of operation range and accumulate necessary data to develop advanced prediction models. Thus, predictors must adapt to evolving plasma states during this exploration phase. To address these challenges, this study further develops the enhanced convolutional autoencoder anomaly detection (E-CAAD) predictor and proposes a cross-tokamak adaptive transfer method based on E-CAAD. By training the E-CAAD model on data from existing devices, the predictor can effectively distinguish between disruption precursor and non-disruption samples in new device, enabling disruption prediction from the first shot on the new device. Additionally, adaptive learning from scratch and alarm threshold adaptive adjustment strategies are proposed to enable model automatically adapt to changes in the discharge scenario. The adaptive learning strategy enables the predictor to fully use scarce data during the early operation of the new device while rapidly adapting to changes in the discharge scenario. The threshold adaptive adjustment strategy addresses the challenge of selecting alarm thresholds on new devices where the validation set is lacking, ensuring that the alarm thresholds adapt to changes in the discharge scenario. Finally, the experiment transferring the model from J-TEXT to EAST exhibit that this method enables disruption prediction from the first shot on EAST, allowing the predictor to adapt to changes in the discharge scenario and maintain high prediction performance.

A deep learning-based disruption prediction algorithm has been implemented on a new tokamak, HL-3. An Area Under receiver-operator characteristic Curve of 0.940 has been realized offline over a test campaign involving 72 disruptive and 240 non-disruptive shots, despite the limited training data available from the initial two campaigns. In addition to the well-documented challenge of insufficient training data, a previously unanticipated issue is addressed that the data distribution of a new device is continuously drifting. The plasma scans across a broad parameter space, bringing a drifting distribution of disruption causes and diagnostic data. This problem is often overlooked in previous implementations on steadily operating tokamaks, necessitating greater attention in future tokamaks like ITER. To address these challenges, innovative modules including predict-first neural network, data augmentation, and pseudo data placeholders are developed and implemented, which promotes the accuracy by up to 20%. A series of advantages are also brought by the modules, including the robustness in handling missing input channels, and the interpretability to identify which parameter of plasma is under abnormal condition. The results demonstrate that, with dedicated data collection and algorithm implementation, the issues of limited data and drifting distribution can be overcome, and further, the deep learning-based algorithm can reliably provide disruption alarms on a new tokamak.

This article describes the use of privileged information to train supervised classifiers, applied for the first time to the prediction of disruptions in tokamaks. The objective consists of making predictions with real-time signals during the discharges (as usual) but after training the predictor also with any kind of data at training time that is not available during discharge execution. The latter kind of data is known as privileged information. Taking into account the limited number of foreseen real time signals for disruption prediction at the beginning of operation in JT-60SA, a predictor with a line integrated density signal and the mode lock signal as privileged information has been developed and tested with 1437 JET discharges. The success rate with positive warning time has been improved from 45.24% to 90.48% and the tardy detection rate has diminished from 50% to 8.33%. The use of privileged information in an adaptive way also provides a remarkable reduction of false alarms from 11.53% to 1.15%. The potential of the methodology, exemplified with data relevant to the beginning of JT-60SA operation, is absolutely general and can be applied to any combination of diagnostic signals.

The high acquisition cost and the significant demand for disruptive discharges for data-driven disruption prediction models in future tokamaks pose an inherent contradiction in disruption prediction research. In this paper, we demonstrated a novel approach to predict disruption in a future tokamak using only a few discharges based on domain adaptation (DA). The approach aims to predict disruption by finding a feature space that is universal to all tokamaks. The first step is to use the existing understanding of physics to extract physics-guided features from the diagnostic signals of each tokamak, called physics-guided feature extraction (PGFE). The second step is to align a few data from the future tokamak (target domain) and a large amount of data from existing tokamaks (source domain) based on a DA algorithm called CORrelation ALignment (CORAL). It is the first attempt at applying DA in the cross-tokamak disruption prediction task. PGFE has been successfully applied in J-TEXT to predict disruption with excellent performance. PGFE can also reduce the data volume requirements due to extracting the less device-specific features, thereby establishing a solid foundation for cross-tokamak disruption prediction. We have further improved CORAL called supervised CORAL (S-CORAL) to enhance its appropriateness in feature alignment for the disruption prediction task. To simulate the existing and future tokamak case, we selected J-TEXT as the existing tokamak and EAST as the future tokamak, which has a large gap in the ranges of plasma parameters. The utilization of the S-CORAL improves the disruption prediction performance on future tokamak. Through interpretable analysis, we discovered that the learned knowledge of the disruption prediction model through this approach exhibits more similarities to the model trained on large data volumes of future tokamak. This approach provides a light, interpretable and few data-required ways by aligning features to predict disruption using small data volume from the future tokamak.

Unmitigated disruptions pose a much more serious threat when large-scale tokamaks are operating in the high performance regime. Machine learning based disruption predictors can exhibit impressive performance. However, their effectiveness is based on a substantial amount of training data. In future reactors, obtaining a substantial amount of disruption data in high performance regimes without risking damage to the machine is highly improbable. Using machine learning to develop disruption predictors on data from the low performance regime and transfer them to the high performance regime is an effective solution for a large reactor-sized tokamak like ITER and beyond. In this study, a number of models are trained using different subsets of data from the HL-2A tokamak experiment. A SHapley Additive exPlanations (SHAP) analysis is executed on the models, revealing that there are different, even contradicting, patterns between different performance regimes. Thus, simply mixing data among different performance regimes will not yield optimal results. Based on this analysis, we propose an instance-based transfer learning technique which trains the model using a dataset generated with an optimized strategy. The strategy involves instance and feature selection based on the physics behind differences in high- and low-performance discharges, as revealed by SHAP model analysis. The TrAdaBoost technique significantly improved the model performance from 0.78 BA (balanced accuracy) to 0.86 BA with a few high-performance operation data.

Reliable disruption prediction (DP) and disruption mitigation systems are considered unavoidable during international thermonuclear experimental reactor (ITER) operations and in the view of the next fusion reactors such as the DEMOnstration Power Plant (DEMO) and China Fusion Engineering Test Reactor (CFETR). In the last two decades, a great number of DP systems have been developed using data-driven methods. The performance of the DP models has been improved over the years both for a more appropriate choice of diagnostics and input features and for the availability of increasingly powerful data-driven modelling techniques. However, a direct comparison among the proposals has not yet been conducted. Such a comparison is mandatory, at least for the same device, to learn lessons from all these efforts and finally choose the best set of diagnostic signals and the best modelling approach. A first effort towards this goal is made in this paper, where different DP models will be compared using the same performance indices and the same device. In particular, the performance of a conventional Multilayer Perceptron Neural Network (MLP-NN) model is compared with those of two more sophisticated models, based on Generative Topographic Mapping (GTM) and Convolutional Neural Networks (CNN), on the same real time diagnostic signals from several experiments at the JET tokamak. The most common performance indices have been used to compare the different DP models and the results are deeply discussed. The comparison confirms the soundness of all the investigated machine learning approaches and the chosen diagnostics, enables us to highlight the pros and cons of each model, and helps to consciously choose the approach that best matches with the plasma protection needs.

Plasma disruption is a very dangerous event for future tokamaks and fusion reactors. Therefore, predicting disruption is crucial for ensuring the safety and performance of reactors. In this study, the features of two deep learning algorithms are integrated to establish a multi-scale hybrid network disruption predictor. Firstly, 43 diagnostic signals are extracted by a convolutional neural network (CNN), and the time information is learned by a long short-term memory network. The predictor is trained and tested on a database containing ${\sim}10^4$ non-all-metal wall discharges. Its area under the receiver operator characteristic curve (AUC), which is a common performance metric for deep learning algorithms, reaches 0.97, and the true positive rate is ${\sim}95.3\%$ , while the false positive rate is ${\sim}8\%$ . Since EAST was upgraded in 2020, the wall condition has been upgraded from non-all-metal to all-metal. To examine the robustness of the predictor, the EAST disruption predictor is migrated to the all-metal wall experiment for the first time. It is again tested under the all-metal wall experimental data, and its warning performance decreases significantly, with an AUC of only 0.79. To improve the robustness and sensitivity of the predictor against disruptions, the convolutional attention mechanism is introduced into the CNN. After training and testing with the same data set, the warning performance for the all-metal wall data is improved, with the AUC value increasing to 0.84. To further improve the robustness of the predictor, the t-distributed stochastic neighbor embedding algorithm is employed to explore the difference of the sample distribution before and after EAST upgradation, and the transfer learning algorithm is used to reduce this difference. By applying transfer learning to a small amount of discharges from the all-metal wall data, the warning performance in the all-metal wall experiment is further improved, and the AUC value increases to 0.93. Overall, the proposed data-driven predictor is promising for applications in future devices such as ITER.

The ability to identify underlying disruption precursors is key to disruption avoidance. In this paper, we present an integrated deep learning (DL) based model that combines disruption prediction with the identification of several disruption precursors like rotating modes, locked modes, H-to-L back transitions and radiative collapses. The first part of our study demonstrates that the DL-based unstable event identifier trained on 160 manually labeled DIII-D shots can achieve, on average, 84% event identification rate of various frequent unstable events (like H-L back transition, locked mode, radiative collapse, rotating MHD mode, large sawtooth crash), and the trained identifier can be adapted to label unseen discharges, thus expanding the original manually labeled database. Based on these results, the integrated DL-based framework is developed using a combined database of manually labeled and automatically labeled DIII-D data, and it shows state-of-the-art (AUC = 0.940) disruption prediction and event identification abilities on DIII-D. Through cross-machine numerical disruption prediction studies using this new integrated model and leveraging the C-Mod, DIII-D, and EAST disruption warning databases, we demonstrate the improved cross-machine disruption prediction ability and extended warning time of the new model compared with a baseline predictor. In addition, the trained integrated model shows qualitatively good cross-machine event identification ability. Given a labeled dataset, the strategy presented in this paper, i.e. one that combines a disruption predictor with an event identifier module, can be applied to upgrade any neural network based disruption predictor. The results presented here inform possible development strategies of machine learning based disruption avoidance algorithms for future tokamaks and highlight the importance of building comprehensive databases with unstable event information on current machines.

Disruption prediction has made rapid progress in recent years, especially in machine learning (ML)-based methods. If a disruption prediction model can be interpreted, it can tell why certain samples are classified as disruption precursors. This allows us to tell the types of incoming disruption for disruption avoidance and gives us insight into the mechanism of disruption. This paper presents a disruption predictor called interpretable disruption predictor based on physics-guided feature extraction (IDP-PGFE) and its results on J-TEXT experiment data. The prediction performance of IDP-PGFE with physics-guided features is effectively improved (true positive rate = 97.27%, false positive rate = 5.45%, area under the ROC curve = 0.98) compared to the models with raw signal input. The validity of the interpretation results is ensured by the high performance of the model. The interpretability study using an attribution technique provides an understanding of J-TEXT disruption and conforms to our prior comprehension of disruption. Furthermore, IDP-PGFE gives a possible mean on inferring the underlying cause of the disruption and how interventions affect the disruption process in J-TEXT. The interpretation results and the experimental phenomenon have a high degree of conformity. The interpretation results also gives a possible experimental analysis direction that the resonant magnetic perturbations delays the density limit disruption by affecting both the MHD instabilities and the radiation profile. PGFE could also reduce the data requirement of IDP-PGFE to 10% of the training data required to train a model on raw signals. This made it possible to be transferred to the next-generation tokamaks, which cannot provide large amounts of data. Therefore, IDP-PGFE is an effective approach to exploring disruption mechanisms and transferring disruption prediction models to future tokamaks.