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.

Survival regression models can achieve longer warning times at similar receiver operating characteristic performance than previously investigated models. Survival regression models are also shown to predict the time until a disruption will occur with lower error than other predictors. Time-to-event predictions from time-series data can be obtained with a survival analysis statistical framework, and there have been many tools developed for this task which we aim to apply to disruption prediction. Using the open-source Auton-Survival package we have implemented disruption predictors with the survival regression models Cox Proportional Hazards, Deep Cox Proportional Hazards, and Deep Survival Machines. To compare with previous work, we also include predictors using a Random Forest binary classifier, and a conditional Kaplan-Meier formalism. We benchmarked the performance of these five predictors using experimental data from the Alcator C-Mod and DIII-D tokamaks by simulating alarms on each individual shot. We find that developing machine-relevant metrics to evaluate models is an important area for future work. While this study finds cases where disruptive conditions are not predicted, there are instances where the desired outcome is produced. Giving the plasma control system the expected time-to-disruption will allow it to determine the optimal actuator response in real time to minimize risk of damage to the device. [ABSTRACT FROM AUTHOR]

Copyright of Southern Energy Construction is the property of Southern Energy Construction Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

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 ∼ 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 ∼ 95.3 % , while the false positive rate is ∼ 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. [ABSTRACT FROM AUTHOR]

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. [ABSTRACT FROM AUTHOR]

Recent experience with metallic devices operating in ITER relevant regions of the operational space, has shown that the disruptivity of these plasmas is unacceptably high. The main causes of the disruptions are linked to impurity accumulation in the core and edge cooling, resulting in unstable current profiles. Avoidance and prevention of the consequent instabilities require the early detection of anomalous electron temperature profiles. A series of indicators have been developed and their performances compared, to find the most suitable inputs for disruption predictors. Their properties are assessed on the basis of information content, reliability and real-time availability. The best performing ones provide much better results than the ones reported in the literature, as shown by both numerical tests with synthetic data and the analysis of experimental signals from JET with the ITER-like wall. They provide better accuracy, lower false alarms and earlier detection. The improved discriminatory capability of the developed indicators is expected to significantly improve the performance of the most advanced predictors recently reported in the literature. [ABSTRACT FROM AUTHOR]

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.

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.

Next generation High Performance (HP) tokamaks risk damage from unmitigated disruptions at high current and power. Achieving reliable disruption prediction for a device’s HP operation based on its Low Performance (LP) data is a key to its success. In this letter, through explorative data analysis and dedicated numerical experiments on multiple existing tokamaks, we demonstrate how the operational regimes of tokamaks can affect the power of a trained disruption predictor. First, our results suggest data-driven disruption predictors trained on abundant LP discharges work poorly on the HP regime of the same tokamak, which is a consequence of the distinct distributions of the tightly correlated signals related to disruptions in these two regimes. Second, we find that matching operational parameters among tokamaks strongly improves cross-machine accuracy which implies our model learns from the underlying scalings of dimensionless physics parameters like q 95, β p and confirms the importance of these parameters in disruption physics and cross machine domain matching from the data-driven perspective. Finally, our results show in the absence of HP data from the target devices, the best predictivity of the HP regime for the target machine can be achieved by combining LP data from the target with HP data from other machines. These results provide a possible disruption predictor development strategy for next generation tokamaks, such as ITER and SPARC, and highlight the importance of developing baseline scenario discharges of future tokamaks on existing machines to collect more relevant disruptive data. [ABSTRACT FROM AUTHOR]

• A disruption prediction system for JET, based on a Convolutional Neural Network, is designed. • The contribution of the Mirnov coil measurement spectrogram to disruption prediction is investigated. • The spatiotemporal information from temperature, radiation, and density profiles and the MHD spectrogram are given as input to the predictor as images. • The model is tested on a dataset different from that used during the training phase. • The predictor performance of the proposed predictor is compared with the same predictor model and the same database trained without Mirnov measurement contribute. The present research focuses on investigating deep neural networks techniques for predicting plasma disruptions in tokamaks. For this purpose, various deep-learning predictive models and several plasma diagnostics will be analyzed, using data gathered during the experimental campaigns conducted at the JET nuclear fusion tokamak between 2011 and 2020. The primary objective of this study is to show that the contribution of the MHD spectrograms increases the disruption predictor performance. The final deep-learning prediction model leverages the capability of Convolutional Neural Networks to directly learn important spatiotemporal information from 1D plasma profiles of temperature, density and power radiation, obtained from High Resolution Thomson Scattering and Bolometer diagnostics, as well as from spectrograms generated by a set of fast magnetic pick-up coils known as Mirnov coils. The Convolutional Neural Network eliminates the need for manual feature extraction methods that characterize the majority of machine learning methods. Using plasma profiles information allows to distinguish between core radiation caused by impurity accumulations and outboard radiation phenomena. Likewise, the decision to incorporate spectrograms from Mirnov coils is based on the diagnostic's ability to measure magnetic fluctuations originating from MHD instabilities, which can lead to disruptions. In order to address phenomena characterized by fast temporal dynamics, the inclusion of the locked mode signal was chosen. This signal is commonly employed at JET to trigger mitigation actions. It is integrated into an alarm scheme that employs both AND/OR logic and optimized thresholds, ensuring its effectiveness. The proposed predictor exhibits significant performance, with only one missed alarm out of 92 disrupted discharges and three false alarms out of 131 regularly terminated discharges in the test set. [ABSTRACT FROM AUTHOR]

A robust disruption prediction system is mandatory in a Tokamak control system as the disruption can cause malfunctioning of the plasma-facing components and impair irrecoverable structural damage to the vessel. To mitigate the disruption, in this article, a data-driven based disruption predictor is developed using an ensemble technique. The ensemble algorithm classifies disruptive and non-disruptive discharges in the GOLEM Tokamak system. Ensemble classifiers combine the predictive capacity of several weak learners to produce a single predictive model and are utilized both in supervised and unsupervised learning. The resulting final model reduces the bias, minimizes variance and is unlikely to over-fit when compared to the individual model from a single algorithm. In this paper, popular ensemble techniques such as Bagging, Boosting, Voting, and Stacking are employed on the time-series Tokamak dataset, which consists of 117 normal and 70 disruptive shots. Stacking ensemble with REPTree (Reduced Error Pruning Tree) as a base learner and Multi-response Linear Regression as meta learner produced better results in comparison to other ensembles. A comparison with the widely employed stand-alone machine learning algorithms and ensemble algorithms are illustrated. The results show the excellent performance of the Stacking model with an F1 score of 0.973. The developed predictive model would be capable of warning the human operator with feedback about the feature(s) causing the disruption. [ABSTRACT FROM AUTHOR]

• Disruptions are a priority research subject in sights of ITER. • It is urgent to find solutions to avoid or at least mitigate their effects. • A fundamental pre-requisite to intervene is to predict their occurrence. • Machine learning techniques can aid to model accurate disruption predictors. • A linear equation derived from a machine learning disruption predictor has been derived. The implementation of Machine Learning techniques can help to develop accurate disruption predictors. However, they may provide outcomes difficult to understand from a physics point of view due to their mathematical formulation. In this work an interpretable linear equation has been derived from a data-driven model. This developed predictor can be used for real-time forecasting and for the off-line analysis of the variables that contribute to the alarm triggering. To create the linear model, in addition to physic quantities, Time Increments (TIs) that represent the differences in the amplitude of a signal divided by a predefined time interval have been considered. To select the best subset of quantities to include (from the wide range of possible combinations of signals and TIs), Genetic Algorithms have been applied. The results, obtained over an independent testing database of 131 unintentional disruptive and 1310 non-disruptive shots, are 99,24% of success rate (94,66% of them with at least 10 ms of warning time) and 3,51% of false alarms. [ABSTRACT FROM AUTHOR]

Disruptions are dangerous events in tokamaks that require mitigation methods to alleviate its detrimental effects. A prerequisite to trigger any mitigation action is the existence of a reliable disruption predictor. This article assesses a predictor that relates in a linear way consecutive samples of a single quantity (in particular, the magnetic perturbation time derivative signal has been used). With this kind of predictor, the recognition of disruptions does not depend on how large the signal amplitude is but on how large the signal increments are: small increments mean smooth plasma evolution whereas abrupt increments reflect a non-smooth evolution and potential risk of disruption. Results are presented with data from the JT-60U tokamak and high-beta discharges. Two training methods have been tested: a classical approach in which the more data for training the better and an adaptive method that starts from scratch. In both cases the success rate is about 95%. It should be noted that predictors based on signal increments and their adaptive versions can be of big interest for next devices such as JT-60SA or ITER. [ABSTRACT FROM AUTHOR]

The ability and flexibility of the open computing language (OpenCL) for task parallelization in heterogeneous computing platforms [field-programmable gate array (FPGA), CPU, and GPU] are remarkable advantages when designing advanced data acquisition and processing systems. The use of FPGA devices in data acquisition devices increases the capabilities of traditional data acquisition systems, thus allowing the implementation of complex mathematical algorithms with the acquired data. This paper shows an OpenCL implementation of an adaptive probabilistic disruption predictor for fusion devices implemented in a Cyclone V SoC device with an ARM processor, the FPGA logic, and an analog to digital converter. This paper describes the methodology used, the hardware/software system architecture, and the implementation results. The work highlights the critical aspects involved in designing these OpenCL-based systems and, in particular, the implementation of the board support package. This paper also presents the aspects such as the significant differences in the design flow concept between FPGA and GPU OpenCL implementations and how to optimize the FPGA implementation with OpenCL. The results show that it is possible to make predictions with computation times shorter than $500~\mu \text{s}$ when using this low-cost SoC. [ABSTRACT FROM AUTHOR]

• Data driven disruption prediction using random forest algorithm is studied in KSTAR. • The model shows an overall accuracy of 94.7 % in both phases, and the F-1 score is 0.93. • The prediction model will be implemented in the KSTAR plasma control system to trigger a disruption mitigation system to reduce the harmful effects of plasma disruption. Here, we develop a data driven disruption prediction model using the random forest algorithm using a database of KSTAR disruption shots. A total of 719 disruption shots from the past 3 campaigns have been selected to generate a database, which are terminated during the plasma current flattop phase by plasma physics issues and have all the essential diagnostic data. To train and test our machine learning model, the input features known to be associated with plasma disruption were collected from various diagnostic and analysis data. The slice datasets were labeled as 'disruption' less than 40 ms from thermal quench, and the remainder as 'non-disruption'. The random forest-based disruption prediction model was trained and tested against the database. An out-of-bag (OBB) error indicates the optimal value of the 'number of trees' in the random forest model. A confusion matrix with test database shows an overall accuracy of 91 % in both phases, and an F-1 score of 0.915. This model shows the characteristics of disruptions that occur in KSTAR, and it can be helpful for disruption prediction studies of future superconducting tokamaks. [ABSTRACT FROM AUTHOR]

Highlights • Disruption predictor based on high-beta plasma experimental data in JT-60U. • Input plasma parameters are selected by sparse modeling using exhaustive search. • Performance of prediction is improved by extracting input parameters. • Key parameters including magnetic fluctuation and its time delivered are extracted. • False alarm rate is improved considering temporal change of parameters. Abstract Disruption is a critical phenomenon in a tokamak reactor. Although disruption causes serious damage to the reactor, its physical mechanism remains unclear. To realize a tokamak reactor, it is necessary to understand and control the disruption phenomenon. The present research constructs a disruption predictor using experimental high-beta plasma data in the JT-60U tokamak. The predictor was constructed using a support vector machine as a linear discriminant, and we focus on a variable selection problem for the binary classification by sparse modeling, specifically, exhaustively searching the best combinations of variables which maximize the predictor performance. By the sparse modeling, we found that the six input parameters as the best combinations. The selected parameters were the n = 1 mode amplitude | B r n = 1 | and its time derivative d | B r n = 1 | / d t , the plasma density (relative to the Greenwald density limit) and its time derivative, and the time derivatives of the plasma internal inductance and plasma elongation. In particular, it was identified that the parameter d | B r n = 1 | / d t , plays a key role on plasma disruption. We should notice that the combination with other plasma parameters is indispensable and remarkably make it possible to improve the performance of disruption prediction. [ABSTRACT FROM AUTHOR]

Abstract The aim of this paper is to present a signal processing algorithm that, applied to the raw Locked Mode signal, allows us to obtain a disruption indicator in principle exploitable on different tokamaks. A common definition of such an indicator for different machines would facilitate the development of portable systems for disruption prediction, which is becoming of increasingly importance for the next tokamak generations. Moreover, the indicator allows us to overcome some intrinsic problems in the diagnostic system such as drift and offset. The behavior of the proposed indicator as disruption predictor, based on crossing optimized thresholds of the signal amplitude, has been analyzed using data of both JET and ASDEX Upgrade experiments. A thorough analysis of the disruption prediction performance shows how the indicator is able to recover some missed and tardy detections of the raw signal. Moreover, it intervenes and corrects premature or even wrong alarms due to, e.g., drifts and/or offsets. [ABSTRACT FROM AUTHOR]

Reliable algorithms for disruption avoidance and prediction are foreseen to play a fundamental role in the JET control system for the successful operation of the machine in the upcoming deuterium-tritium campaigns. The integration of such algorithms is expected to be a key part also in the implementation of the ITER plasma control system. So far, most of the effort has been devoted to the prediction of disruptions, which is required to mitigate the effects of these transient events, protecting the integrity of in-vessel components. Nevertheless, in order to put in place recover strategies or to have the possibility of a soft landing for the plasma current, the paradigm must be shifted to avoiding disruptions. In this paper, plasma profile-based indicators will be statistically analyzed showing their potential in such a perspective, where warning times and reliability of detection are crucial. [ABSTRACT FROM AUTHOR]

Real-time (RT) disruption prediction (DP) is essential to trigger mitigation actions that avoid irreversible damage to the devices. This paper deals with disruption mitigation alarms and performs the RT implementation of a probabilistic predictor. The RT implementation has been carried out with a fast controller with DAQ FPGA-based data acquisition devices corresponding to ITER catalogue (in particular, a reconfigurable Input/Output platform has been used). Up to three input signals have been used and relevant information for the prediction is extracted from the temporal and the frequency domains. The signals are read from the JET database. Then D/A conversions are carried out and used as inputs to the real time system. In this way, the whole process of digitization, data analysis and prediction is performed. The computation time for each prediction takes less than 200 μs. [ABSTRACT FROM AUTHOR]

Plasma disruptions are a serious impediment to tokamak operations, and accordingly there has been a lot of attention given to the topic of predicting disruptions in order to avoid or more effectively mitigate them. While existing prediction goals are based on a combination of engineering constraints and desired operating outcomes, it does not follow that a predictor must meet these strict requirements to offer a benefit. Rather, a cost-based criterion should provide the necessary justification to implement a disruption predictor. A general model for such a cost-based criterion is derived here. This model is used to demonstrate that, with realistic assumptions about the cost of disruptions compared to the cost of unnecessary discharge terminations, existing disruption prediction techniques already meet the minimum criterion to reduce operating costs. [ABSTRACT FROM AUTHOR]

Since year 2010, the APODIS architecture has proven its accuracy predicting disruptions in JET tokamak. Nevertheless, it has shown margins for improvements, fact indisputable after the enhanced performances achieved in posterior upgrades. In this article, a complete optimization driven by Genetic Algorithms (GA) is applied to it aiming at considering all possible combination of signals, signal features, quantity of models, their characteristics and internal parameters. This global optimization targets the creation of the best possible system with a reduced amount of required training data. The results harbor no doubts about the reliability of the global optimization method, allowing to outperform the ones of previous versions: 91.77% of predictions (89.24% with an anticipation higher than 10 ms) with a 3.55% of false alarms. Beyond its effectiveness, it also provides the potential opportunity to develop a spectrum of future predictors using different training datasets. [ABSTRACT FROM AUTHOR]

In large-scale tokamaks disruptions have the potential to create serious damage to the facility. Hence disruptions must be avoided, but, when a disruption is unavoidable, minimizing its severity is mandatory. A reliable detection of a disruptive event is required to trigger proper mitigation actions. To this purpose machine learning methods have been widely studied to design disruption prediction systems at ASDEX Upgrade. The training phase of the proposed approaches is based on the availability of disrupted and non-disrupted discharges. In literature disruptive configurations were assumed appearing into the last 45 ms of each disruption. Even if the achieved results in terms of correct predictions were good, it has to be highlighted that the choice of such a fixed temporal window might have limited the prediction performance. In fact, it generates confusing information in cases of disruptions with disruptive phase different from 45 ms. The assessment of a specific disruptive phase for each disruptive discharge represents a relevant issue in understanding the disruptive events. In this paper, the Mahalanobis distance is applied to define a specific disruptive phase for each disruption, and a logistic regressor has been trained as disruption predictor. The results show that enhancements on the achieved performance on disruption prediction are possible by defining a specific disruptive phase for each disruption. [ABSTRACT FROM AUTHOR]

• A deep learning- based real-time disruption predictor is implemented in the plasma control system (PCS) in HL-2A. It works well during the online testing of 382 shots and gets a TPR of 95.8% and a TNR of 77.5%. • A python interpreter + TFlite paradigm is proposed to implement AI modules in PCS. AI-based algorithms can be implemented into the C/C++ based PCS with little modification. Only 0.06 ms of time consumption and almost no penalty on the algorithm's performance will be introduced during this process. • An integrated system for disruption prediction and mitigation is developed in HL-2A. Some closed-loop online testing is implemented, during which the algorithm successfully triggered the MGI and SMBI system to mitigate the coming disruption. A real-time disruption predictor based on deep learning method is implemented into the Plasma Control System (PCS) of HL-2A. It has a total accuracy of 89.0% during the online testing of Shot Nos. 38,650–39,347 in HL-2A. 32 false alarms are triggered during the 142 non-disruptive shots and 10 disruptions are missed in the 240 disruptive shots. In several shots, this disruption prediction algorithm is used to trigger the disruption mitigation system, namely, massive gas injection (MGI) and supersonic molecular beam injection (SMBI) in this experiment. For MGI, there is a 12-millisecond delay between the trigger signal and the mitigated disruption. In the 240 disruptive shots, 81.5% of them are predicted with more than 12 ms in advance and thus can be successfully mitigated. For SMBI, the delay is 4 millisecond and 91.3% of the disruptions can be successfully mitigated. In general, although dedicated work is needed to close the gap between online disruption prediction and the offline algorithms, the effectiveness of deep learning-based disruption predictors in real-time environments is validated in this research. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, a disruption prediction system for ASDEX Upgrade has been proposed that does not require disruption terminated experiments to be implemented. The system consists of a data-based model, which is built using only few input signals coming from successfully terminated pulses. A fault detection and isolation approach has been used, where the prediction is based on the analysis of the residuals of an auto regressive exogenous input model. The prediction performance of the proposed system is encouraging when it is applied to the same set of campaigns used to implement the model. However, the false alarms significantly increase when we tested the system on discharges coming from experimental campaigns temporally far from those used to train the model. This is due to the well know aging effect inherent in the data-based models. The main advantage of the proposed method, with respect to other data-based approaches in literature, is that it does not need data on experiments terminated with a disruption, as it uses a normal operating conditions model. This is a big advantage in the prospective of a prediction system for ITER, where a limited number of disruptions can be allowed. [Copyright &y& Elsevier]

Highlights: [•] JET real-time disruption predictor with metallic wall 991 discharges analyzed. [•] Predictor training has been carried out with JET C wall data. [•] Success, false alarm and missed alarm rates are 98.4%, 0.9% and 1.6%, respectively. [•] Alarms are triggered in average 426ms before the disruption. [Copyright &y& Elsevier]

The Self-Organizing Map is a computational method for the visualization and analysis of high-dimensional data. Self Organizing Maps have been applied to ASDEX Upgrade data to define an ordered mapping of an 8-dimensional plasma parameter space onto a regular, 2-dimensional grid. The map has been used to track the plasma trajectory during the experiments and monitor the disruption risk. In order to face with ever new operational conditions, a periodical updating of the Self Organizing Map is proposed. [ABSTRACT FROM AUTHOR]

The mapping of the n-dimensional plasma parameter space of ASDEX Upgrade (AUG) has been performed using a 2-D self-organizing map (SOM), which reveals the map potentiality in data visualization. The proposed approach allows us the definition of simple displays capable of presenting meaningful information on the actual state of the plasma, but it also suggests to use the SOM as a disruption predictor. In this paper, various criteria have been studied to associate the risk of disruption of each cluster in the map to a disruption alarm threshold. The data for this study come from AUG experiments executed between July 2002 and November 2009. The prediction performance of the proposed system has been evaluated on a set of discharges different from those used for the map training, obtaining a good prediction success rate. A visual analysis of the predictor input signals has been performed for wrong predictions in order to identify possible common causes, and some criteria to increase prediction performance have been derived. [ABSTRACT FROM PUBLISHER]

Abstract: In this paper, an adaptive neural system has been built to predict the risk of disruption at ASDEX Upgrade. The system contains a Self Organizing Map, which determines the ‘novelty’ of the input of a Multi Layer Perceptron predictor module. The answer of the MLP predictor will be inhibited whenever a novel sample is detected. Furthermore, it is possible that the predictor produces a wrong answer although it is fed with known samples. In this case, a retraining procedure will be performed to update the MLP predictor in an incremental fashion using data coming from both the novelty detection, and from wrong predictions. In particular, a new update is performed whenever a missed alarm is triggered by the predictor. The performance of the adaptive predictor during the more recent experimental campaigns until November 2009 has been evaluated. [Copyright &y& Elsevier]

This paper presents a recently introduced Kernel Machine, called Geometrical Kernel Machine, used to predict disruptive events in nuclear fusion reactors. The algorithm proposed to construct the Kernel Machine is able to automatically determine both the number of neurons and the synaptic weights of a Multilayer Perceptron neural network with a single hidden layer. It has been demonstrated that the resulting network is able to classify any finite set of patterns defined in a real domain. The prediction problem has been here modeled as a two classes classification problem. The geometrical interpretation of the network equations allows us both to develop the disruption predictor and to manage the so called ageing of the kernel machine. In fact, using the same kernel machine, a novelty detection system has been integrated in the predictor, increasing the overall system performance, and the reliability of the predictor. [ABSTRACT FROM AUTHOR]

Abstract: Magnetic confinement fusion devices are very integrated systems, difficult to access for measurement, and therefore they pose particular challenges to identification and to the prediction of undesired events. With regard to system identification, Bayesian statistics is a very promising methodology, which provides for the first time a sound way to include physical information about the diagnostics in the evaluation of the error bars. In a prototypical application, a Bayesian statistical approach determines the magnetic topology without any assumption on the equilibrium and provides clear confidence intervals on all the derived quantities. This technique, which complements another more traditional approach based on the Grad–Shafranov equation, can be implemented to provide results about every few milliseconds and therefore it can be envisaged to exploit for feedback purposes. Some phenomena in the evolution of Tokamak plasmas, like disruptions, are too dangerous and prohibitively difficult to control. For these cases avoidance is the best alternative and therefore specific classifiers have been trained and optimized to predict the occurrence of disruptions. The success rates of these predictors, mainly based on Support Vector Machines, are very often of the order or higher than 90%. The generalisation capability of the method has been confirmed by applying the same predictor to new campaigns without retraining. The success rate remains very high (above 80%) even 12 campaigns after the last one used for training. [ABSTRACT FROM AUTHOR]

Abstract: The present understanding of disruption physics has not gone so far as to provide a mathematical model describing the onset of this instability. A disruption prediction system, based on a statistical analysis of the diagnostic signals recorded during the experiments, would allow estimating the probability of a disruption to take place. A crucial point for a good design of such a prediction system is the appropriateness of the data set. This paper reports the details of the database built to train a disruption predictor based on neural networks for ASDEX Upgrade. The criteria of pulses selection, the analyses performed on plasma parameters and the implemented pre-processing algorithms, are described. As an example of application, a short description of the disruption predictor is reported. [Copyright &y& Elsevier]

Disruptions remain a serious issue for International Thermonuclear Experimental Reactor (ITER) and the commercial reactor. In the past years, significant efforts have been devoted to the development of reliable disruption predictors. In this paper, the potential of fuzzy logic in this field is explored. A new disruption predictor for Joint European Torus (JET), which is based on fuzzy logic, was tested off-line using data collected during several JET campaigns, covering a very wide range of scenarios and plasma parameters. In the standard configuration, it was able to predict 92% of disruptions with about 18% of false alarms (FAs). The flexibility of the approach, which is to provide a different tradeoff between missed and FAs, depending on the desired results in terms of safety, is demonstrated. The potential of the technique to help quantifying the risks that are associated with the operation outside the already explored parameter space is also discussed. [ABSTRACT FROM AUTHOR]

Abstract: In the last years there has been a growing interest on black box approaches to disruption prediction. The drawback of these approaches is that the system could deteriorate its performance once it does not get updated. This could be the case of a disruption predictor for JET, where new plasma configurations might present features completely different from those observed in the experiments used during the training phase. This ‘novelty’ can be incorrectly classified by the system. A novelty detection method, which determines the novelty of the input of the prediction system, can be used to assess the system reliability. This paper presents a support vector machines disruption predictor for JET, wherein multiple plasma diagnostic signals are combined to provide a composite impending disruption warning indicator. In a support vector machine the analysis of the decision function value gives useful information about the novelty of an input and, on the reliability of the predictor output, during on-line applications. Results show the suitability of support vector machines both for prediction and novelty detection tasks at JET. [Copyright &y& Elsevier]

A research area that has become increasingly important in recent years is that of on-board mobile communication, where users on a vehicle are connected to a local network that attaches to the Internet via a mobile router and a wireless link. In this architecture, link disruptions (e.g., due to signal degradation) may have an immediate impact on a potentially large number of connections. We argue that the advance knowledge of public transport routes, and their repetitive nature, allows a certain degree of prediction of impending link disruptions, which can be used to offset their catastrophic impact. Focusing on the transmission control protocol (TCP) and its extension known as Freeze-TCP, we present a detailed analysis of the performance improvement of TCP connections in the presence of disruption prediction. In particular, we propose a Markov model of Freeze-TCP that captures both the TCP behavior and the prediction+"freezing" feature and, using simulations, show that it accurately predicts the performance of the protocol. Our results demonstrate the significant throughput improvement that can be gained by disruption prediction, even with random packet losses or imperfect timing of the predicted disruptions. [ABSTRACT FROM AUTHOR]

• A database built with disruption related information. • This database makes it easy to develop disruption predictors. • The disruption database can be applied to different Tokamak devices. • The process of generating disruption related information can be automated. Machine learning based disruption prediction methods have exhibited good prediction performance with higher success rate, lower false alarm rate and earlier warning time than physical based methods. One important thing pushed recent advances in machine learning field is high-quality training data. So a database with rich set of accurate disruption related information is crucial to the development of a high performance disruption predictor. In order to develop machine learning based disruption predictors fast and iteratively on J-TEXT, a database dedicated for machine learning disruption prediction has been built. This database provides not only disruption related labels, interfaces for querying the data based on user proved filter and auto generation of training and test data, but also interface for benchmark the predictors. Its modular design allows us to plug-in various shot analysis modules which process diagnostic signals and generate different labels automatically. These modules can be scheduled and run parallel on a cluster which will speed up the shot analysis process. The generated labels are inserted into a MongoDB NoSQL database for later querying. But one major hurdle for machine learning disruption prediction is they perform un-acceptably poor on devices other than they are trained on. It requires data from many different tokamaks to possibly develop a cross machine predictor, so this database also has a data import module which reads diagnostic data from different data sources such as MDSplus and store them as HDF5 files with unified data structure on a parallel file system. It is easy to import data from different machines and provide a unified data access interface. [ABSTRACT FROM AUTHOR]