A statistical approach for the automatic identification of the start of the chain of events leading to the disruptions at JET

Abstract

This paper reports an algorithm to automatically identify the chain of events leading to a disruption, evaluating the so-called reference warning time. This time separates the plasma current flat-top of each disrupted discharge into two parts: a non-disrupted part and a pre-disrupted one. The algorithm can be framed into the anomaly detection techniques as it aims to detect the off-normal behavior of the plasma. It is based on a statistical analysis of a set of dimensionless plasma parameters computed for a selection of discharges from the JET experimental campaigns. In every data-driven model, such as the generative topographic mapping (GTM) predictor proposed in this paper, it is indeed necessary to label the samples needed for training the model itself. The samples describing the disruption-free behavior are extracted from the plasma current flat-top phase of the regularly terminated discharges. The disrupted space is described by all the samples belonging to the pre-disruptive phase of each disruptive discharge in the training set. Note that a proper selection of the pre-disruptive phase plays a key role in the prediction performance of the model. Moreover, these models, which are highly dependent on the training input space, may be particularly prone to degradation as the operational space of any experimental machine is continuously evolving. Hence, a regular schedule of model review and retrain must be planned. The proposed algorithm avoids the cumbersome and time-consuming manual identification of the warning times, helping to implement a continuous learning system that could be automated, despite being offline. In this paper, the automatically evaluated warning times are compared with those obtained with a manual analysis in terms of the impact on the mapping of the JET input parameter space using the GTM methodology. Moreover, the algorithm has been used to build the GTM of recent experimental campaigns, with promising results.

Real-time prediction of high-density EAST disruptions using random forest

“…Therefore, a real-time capable algorithm tailored to detect those cases would be highly beneficial to reduce the disruption rate in EAST and improve the experimental performances. Lastly, the choice of a refined τ class for each disruptive discharge, as opposed to the assumption of a unique value, was seen to increase the predictive performances of data-driven algorithms [31,35]. Therefore, as part of our future work, we plan to refine the τ class definition on a shot-by-shot basis by taking advantage of automated classification methods to extract metadata information that have been developed in the meanwhile [36].…”

Disruption prediction based on fusion feature extractor on J-TEXT

“…Some of the traditional machine learning methods have obtained acceptable result on their own devices. JET developed algorithms based on statistical approaches, [8] CART (classification and regression trees) based on ensemble decision trees, [9] and APODIS based on support vector machine (SVM). [10] DIII-D has also developed disruption prediction using random forests (DPRF).…”

Disruption prediction at JET through deep convolutional neural networks using spatiotemporal information from plasma profiles

“…In over two decades of investigations, artificial intelligence-based approaches demonstrated the great potential to predict disruptions in tokamak devices. Several machine learning methods such as multi-layer perceptron neural networks, support vector machines, self organizing maps and generative topographic mapping (GTM), classification and regression trees and random forests have been used to develop disruption prediction models for JET [2][3][4][5][6][7][8], ASDEX Upgrade [9][10][11], EAST [12], J-TEXT [13], DIIID [14] and Alcator C-Mod [12]. All these contributions have highlighted the importance of studying in depth the physical phenomena involved in disruptive processes in order to synthesize suitable disruption precursor signals to be used as inputs in the data-driven prediction models for avoidance or mitigation actions [7,[15][16][17][18].…”

“…To this end, 0D peaking factor signals have been introduced to encode the spatial information contained in the 1D profiles, through the ratio between the mean values of measurements over different regions of the plasma cross section. The peaking factor signals, constructed starting from temperature, density and plasma radiation profiles, and therefore well anchored to the plasma physics, demonstrated to increase the performance of the machine learning models predicting disruptions with enough warning time to more efficiently enable avoidance strategies [7,8,15,17]. As an example, in [7,15], the peaking factors of the radial profiles of temperature and density are defined as the ratio between the mean value around the magnetic axis and the mean value of the measurements over the entire radius.…”

“…Its prediction performance has been evaluated using disrupted and regularly terminated discharges from a decade of JET experimental campaigns, from 2011 to 2020, showing the robustness of the algorithm, even though the operational spaces in the different campaigns may change significantly. Moreover, the performance of the proposed disruption predictor has been compared, on the same test sets, with the performance of the GTM predictor [7,8] implemented in the Plasma Event TRiggering and Alarms (PETRA) system at JET [29], with even better results. Furthermore, monitoring the deep-CNN predictor output together with the provided input images, it is possible to interpret the physics mechanisms leading to the model response.…”