Abstract:The inadequacies of basic physics models for disruption prediction have induced the community to increasingly rely on data mining tools. In the last decade, it has been shown how machine learning predictors can achieve a much better performance than those obtained with manually identified thresholds or empirical descriptions of the plasma stability limits. The main criticisms of these techniques focus therefore on two different but interrelated issues: poor “physics fidelity” and limited interpretability. Insu… Show more
“…The equation has been particularised using the same campaigns of the cited article [47] with exactly the same setup. To this end, first the disruption probability is modelled with the sigmoid function reported in equation (2.2):…”
Section: Detecting Anomalies In the Magnetic Configurationmentioning
confidence: 99%
“…Hybrid physics and data-driven training. A few works on how to scale anomaly indicators between machines have been published [47,48]. Their scope is quite limited though.…”
Section: A Data-driven Physics-based Approach To Prediction For Proxi...mentioning
Disruptions are abrupt collapses of the configuration that have afflicted all tokamaks ever operated. Reliable observers are a prerequisite to the definition and the deployment of any realistic strategy of countermeasures to avoid or mitigate disruptions. Lacking first principle models of the dynamics leading to disruptions, in the past decades empirical predictors have been extensively studied and some were even installed in JET real-time network. Having been conceived as engineering tools, they were often very abstract. In this work, physics and data-driven methodologies are combined to identify the main macroscopic precursors of disruptions: magnetic instabilities, abnormal kinetic profiles and radiation patterns. Machine learning predictors utilising these observers can not only detect and classify these anomalies but also determine their probability of occurrence and estimate the time remaining before their onset. These tools have been applied to a database of about two thousand JET discharges with various isotopic compositions including DT, in conditions simulating in all respects real time deployment. Their performance would meet ITER requirements, and they are expected to be easily transferrable to larger devices, because they rely only on normalised quantities, form factors, and physical/empirical scaling laws.
“…The equation has been particularised using the same campaigns of the cited article [47] with exactly the same setup. To this end, first the disruption probability is modelled with the sigmoid function reported in equation (2.2):…”
Section: Detecting Anomalies In the Magnetic Configurationmentioning
confidence: 99%
“…Hybrid physics and data-driven training. A few works on how to scale anomaly indicators between machines have been published [47,48]. Their scope is quite limited though.…”
Section: A Data-driven Physics-based Approach To Prediction For Proxi...mentioning
Disruptions are abrupt collapses of the configuration that have afflicted all tokamaks ever operated. Reliable observers are a prerequisite to the definition and the deployment of any realistic strategy of countermeasures to avoid or mitigate disruptions. Lacking first principle models of the dynamics leading to disruptions, in the past decades empirical predictors have been extensively studied and some were even installed in JET real-time network. Having been conceived as engineering tools, they were often very abstract. In this work, physics and data-driven methodologies are combined to identify the main macroscopic precursors of disruptions: magnetic instabilities, abnormal kinetic profiles and radiation patterns. Machine learning predictors utilising these observers can not only detect and classify these anomalies but also determine their probability of occurrence and estimate the time remaining before their onset. These tools have been applied to a database of about two thousand JET discharges with various isotopic compositions including DT, in conditions simulating in all respects real time deployment. Their performance would meet ITER requirements, and they are expected to be easily transferrable to larger devices, because they rely only on normalised quantities, form factors, and physical/empirical scaling laws.
“…The equation of the boundary between disruptive and non-disruptive regions in JET's operational space, with an ITER-like wall as displayed in Fig. 3, was obtained with symbolic regression and revealed information on factors that are likely to trigger disruptions 49 .…”
Section: Interpretable Models With Symbolic Regressionmentioning
“…The physics-driven models try to combine the advantages of both paradigms by adopting surrogate machine learning (ML) models [14], which could also improve the interpretability of these ML models. In data-driven method studies, an interpretability model, achieved by applying symbolic regression methods [15][16][17], has been obtained with the support vector machine in JET. An approach to interpret the 1.5D convolutional neural network model has been developed in HL-2A [18], which is a counterfactual-based interpretable approach.…”
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 (TPR = 97.27%, FPR = 5.45%, AUC = 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 means of 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 give a possible experimental analysis direction that the RMPs delay 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.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.