A linear equation based on signal increments to predict disruptive behaviours and the time to disruption on JET

Abstract

This article describes the development of a generic disruption predictor that is also used as basic system to provide an estimation of the time to disruption at the alarm times. The mode lock signal normalised to the plasma current is used as input feature. The recognition of disruptive/non-disruptive behaviours is not based on a simple threshold of this quantity but on the evolution of the amplitudes between consecutive samples taken periodically. The separation frontier between plasma behaviours (disruptive/non-disruptive) is linear in such parameter space. The percentages of recognised and false alarms are 98% and 4%, respectively. The recognised alarms can be split into valid alarms (90%) and late detections (8%). The experimental distribution of warning times follows an exponential model with average warning time of 443 ms. On the other hand, the prediction of the time to the disruption has been fitted to a Weibull model that relates this predicted time to the distance of the points to the diagonal in the parameter space of consecutive samples. The model shows a very good agreement between predicted times and warning times in narrow time intervals (between 0.01 s and 0.06 s) before the disruption.

On the transfer of adaptive predictors between different devices for both mitigation and prevention of disruptions

“…All the attempts to find new early precursors of disruptions have so far failed, at least on JET, but more efforts are going to be devoted to this task in the future. Spectroscopic measurements [33] possibly using dedicated diagnostics, are a good candidate to allow progress in this direction [34,35]. The bolometric divertor-all profile indicator is calculated as: Bolo Div,All (t) =…”

Disruption prediction with artificial intelligence techniques in tokamak plasmas

“…Of the three machine-learning-based predictors that were implemented in JET's real-time network, APODIS 17 , SPAD 36 and Centroid 37 , the former correctly identified disruptions in more than 98% of cases and had a false alarm rate (that is, wrongly classified non-disruptive data) in fewer than 2% of cases, with an average warning time of hundreds of milliseconds.…”

“…Even predictors with a good performance in terms of success and false alarm rates suffer from a major limitation: on JET, the range of their warning times can be of the order of 1 s. Thus, the control system lacks the information about the time remaining before the beginning of the current quench-it could be in a few milliseconds or in a second. Given the importance of predicting the time remaining before the occurrence of a disruption 37 or at least providing a robust estimate of the minimum time still available to introduce remedial actions 50,51 , accurate estimates of the warning time are indispensable. A recent study on JET combining support vector machines and genetic programming 52 allowed the integration of three classes of predictor, for the avoidance, prevention and mitigation of disruptions.…”

PHAD: a phase-oriented disruption prediction strategy for avoidance, prevention, and mitigation in JET

“…To address the drawback of not being able to complete prevention measures due to the lack of estimations of the time to the disruption, research on predicting the time to the disruption is more than welcome. A recent approach to this [8] presented promising results, but it is only a starting point.…”

“…The robustness of APODIS was verified in practice over the years; its prediction rates remained high, without any retraining of the model, even after considerable structural modifications of the device such as the installation of a new ITER-like wall. Another two prediction models worth mentioning are based on centroid methods [8] and outlier detections [14,15]. This last model (SPAD predictor) achieved high recognition rates in JET by identifying abnormal changes in the locked mode signal.…”