Giuliana Sias

An adaptive real-time disruption predictor for ASDEX Upgrade

In this paper, a neural predictor has been built using plasma discharges selected from two years of ASDEX Upgrade experiments, from July 2002 to July 2004. In order to test the real-time prediction capability of the system, its performance has been evaluated using discharges coming from different experimental campaigns, from June 2005 to July 2007. All disruptions that occurred in the chosen experimental campaigns were included with the exception of those occurring in the ramp-up phase, in the ramp-down phase (if the disruption does not happen in the first 100 ms), those caused by massive gas injection and disruptions following vertical displacement events. The large majority of selected disruptions are of the cooling edge type and typically preceded by the growth of tearing modes, degradation of the thermal confinement and enhanced plasma radiation. A very small percentage of them happen at large beta after a short precursor phase. For each discharge, seven plasma diagnostic signals have been selected from numerous signals available in real-time. During the training procedure, a self-organizing map has been used to reduce the database size in order to improve the training of the neural network. Moreover, an optimization procedure has been performed to discriminate between safe and pre-disruptive phases. The prediction success rate has been further improved, performing an adaptive training of the network whenever a missed alarm is triggered by the predictor.

Automatic disruption classification based on manifold learning for real-time applications on JET

Disruptions remain the biggest threat to the safe operation of tokamaks. To efficiently mitigate the negative effects, it is now considered important not only to predict their occurrence but also to be able to determine, with high probability, the type of disruption about to occur. This paper reports the results obtained using the nonlinear generative topographic map manifold learning technique for the automatic classification of disruption types. It has been tested using an extensive database of JET discharges selected from JET campaigns from C15 (year 2005) up to C27 (year 2009). The success rate of the classification is extremely high, sometimes reaching 100%, and therefore the prospects for the deployment of this tool in real time are very promising.

Tracking of the plasma states in a nuclear fusion device using SOMs

Knowledge discovery consists of finding new knowledge from databases where dimension, complexity, or amount of data is prohibitively large for human observation alone. The need for efficient data visualization and clustering is often faced, for instance, in the analysis, monitoring, fault detection, or prediction of various engineering plants. In this paper, two clustering techniques, K-means and Self-Organizing Maps, are used for the identification of characteristic regions for plasma scenario in nuclear fusion experimental devices. The choice of the number of clusters, which heavily affects the performance of the mapping, is firstly faced. Then, the ASDEX Upgrade Tokamak high-dimensional operational space is mapped into lower-dimensional maps, allowing to detect the regions with high risk of disruption, and, finally, the current process state and its history in time are visualized as a trajectory on the Self-Organizing Map, in order to predict the safe or disruptive state of the plasma.

Manifold learning to interpret JET high-dimensional operational space

In this paper, the problem of visualization and exploration of JET high-dimensional operational space is considered. The data come from plasma discharges selected from JET campaigns from C15 (year 2005) up to C27 (year 2009). The aim is to learn the possible manifold structure embedded in the data and to create some representations of the plasma parameters on low-dimensional maps, which are understandable and which preserve the essential properties owned by the original data.A crucial issue for the design of such mappings is the quality of the dataset. This paper reports the details of the criteria used to properly select suitable signals downloaded from JET databases in order to obtain a dataset of reliable observations. Moreover, a statistical analysis is performed to recognize the presence of outliers. Finally data reduction, based on clustering methods, is performed to select a limited and representative number of samples for the operational space mapping.The high-dimensional operational space of JET is mapped using a widely used manifold learning method, the self-organizing maps. The results are compared with other data visualization methods. The obtained maps can be used to identify characteristic regions of the plasma scenario, allowing to discriminate between regions with high risk of disruption and those with low risk of disruption.

Tracking of the Plasma States in a Nuclear Fusion Device using SOMs

Knowledge discovery consists of finding new knowledge from data bases where dimension, complexity or amount of data is prohibitively large for human observation alone. The Self Organizing Map (SOM) is a powerful neural network method for the analysis and visualization of high-dimensional data. The need for efficient data visualization and clustering is often faced, for instance, in the analysis, monitoring, fault detection, or prediction of various engineering plants. In this paper, the use of a SOM based method for prediction of disruptions in experimental devices for nuclear fusion is investigated. The choice of the SOM size is firstly faced, which heavily affects the performance of the mapping. Then, the ASDEX Upgrade Tokamak high dimensional operational space is mapped onto the 2-dimensional SOM, and, finally, the current process state and its history in time has been visualized as a trajectory on the map, in order to predict the safe or disruptive state of the plasma.

Overview of manifold learning techniques for the investigation of disruptions on JET

Identifying a low-dimensional embedding of a high-dimensional data set allows exploration of the data structure. In this paper we tested some existing manifold learning techniques for discovering such embedding within the multidimensional operational space of a nuclear fusion tokamak. Among the manifold learning methods, the following approaches have been investigated: linear methods, such as principal component analysis and grand tour, and nonlinear methods, such as self-organizing map and its probabilistic variant, generative topographic mapping. In particular, the last two methods allow us to obtain a low-dimensional (typically two-dimensional) map of the high-dimensional operational space of the tokamak.These maps provide a way of visualizing the structure of the high-dimensional plasma parameter space and allow discrimination between regions characterized by a high risk of disruption and those with a low risk of disruption. The data for this study comes from plasma discharges selected from 2005 and up to 2009 at JET. The self-organizing map and generative topographic mapping provide the most benefits in the visualization of very large and high-dimensional datasets. Some measures have been used to evaluate their performance. Special emphasis has been put on the position of outliers and extreme points, map composition, quantization errors and topological errors.

Autosomal dominant nocturnal frontal lobe epilepsy seizure characterization through wavelet transform of eeg records and self organizing maps

In this paper, a Manifold Learning approach for the automatic detection of Autosomal Dominant Nocturnal Frontal Lobe Epilepsy seizures is presented, with the aim to support neurologists in the labelling efforts. Features extracted from polysomnography signals are used in order to detect and discriminate seizure epochs. This task has been addressed by mapping the electroencephalographic signal epochs in different regions of the features space. The result is a Self Organizing Map, which allows to investigate over not straightforward relations in the complex input space for the characterization of seizures.

Automatic disruption classification in JET with the ITER-like wall

The new full-metal ITER-like wall at JET was found to have a deep impact on the physics of disruptions at JET. In order to develop disruption classification, the 10D operational space of JET with the new ITER-like wall has been explored using the generative topographic mapping method. The 2D map has been exploited to develop an automatic disruption classification of several disruption classes manually identified. In particular, all the non-intentional disruptions have been considered, that occurred in JET from 2011 to 2013 with the new wall. A statistical analysis of the plasma parameters describing the operational spaces of JET with carbon wall and JET ITER-like wall has been performed and some physical considerations have been made on the difference between these two operational spaces and the disruption classes which can be identified. The performance of the JET- ITER-like wall classifier is tested in real-time in conjunction with a disruption predictor presently operating at JET with good results. Moreover, to validate and analyse the results, another reference classifier has been developed, based on the k-nearest neighbour technique. Finally, in order to verify the reliability of the performed classification, a conformal predictor based on non-conformity measures has been developed.

Adaptive mapping of the plasma operational space of ASDEX Upgrade for disruption prediction

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 aregular, 2-dimensional grid. The map has been used to trackthe plasma trajectory during theexperiments andmonitorthedisruptionrisk. Inordertofacewithevernewoperationalconditions, aperiodicalupdatingoftheSelfOrganizing Map is proposed.

Convolutional Neural Networks for the Identification of Filaments from Fast Visual Imaging Cameras in Tokamak Reactors

The paper proposes a region-based deep learning convolutional neural network to detect objects within images able to identify the filamentary plasma structures that arise in the boundary region of the plasma in toroidal nuclear fusion reactors. The images required to train and test the neural model have been synthetically generated from statistical distributions, which reproduce the statistical properties in terms of position and intensity of experimental filaments. The recently proposed Faster Region-based Convolutional Network algorithm has been customized to the problem of identifying the filaments both in location and size with the associated score. The results demonstrate the suitability of the deep learning approach for the filaments detection.