J. Vega

An advanced disruption predictor for JET tested in a simulated real-time environment

Disruptions are sudden and unavoidable losses of confinement that may put at risk the integrity of a tokamak. However, the physical phenomena leading to disruptions are very complex and non-linear and therefore no satisfactory model has been devised so far either for their avoidance or their prediction. For this reason, machine learning techniques have been extensively pursued in the last years. In this paper a real-time predictor specifically developed for JET and based on support vector machines is presented. The main aim of the present investigation is to obtain high recognition rates in a real-time simulated environment. To this end the predictor has been tested on the time slices of entire discharges exactly as in real world operation. Since the year 2000, the experiments at JET have been organized in campaigns named sequentially beginning with campaign C1. In this paper results from campaign C1 (year 2000) and up to C19 (year 2007) are reported. The predictor has been trained with data from JET’s campaigns up to C7 with particular attention to reducing the number of missed alarms, which are less than 1%, for a test set of discharges from the same campaigns used for the training. The false alarms plus premature alarms are of the order of 6.4%, for a total success rate of more than 92%. The robustness of the predictor has been proven by testing it with a wide subset of shots of more recent campaigns (from C8 to C19) without any retraining. The success rate over the period between C8 and C14 is on average 88% and never falls below 82%, confirming the good generalization capabilities of the developed technique. After C14, significant modifications were implemented on JET and its diagnostics and consequently the success rates of the predictor between C15 and C19 decays to an average of 79%. Finally, the performance of the developed detection system has been compared with the predictions of the JET protection system (JPS). The new predictor clearly outperforms JPS up to about 180 ms before the disruptions.

PXI-based architecture for real time data acquisition and distributed dynamical data processing

This paper describes an architecture model for data acquisition systems based on compact PCI platforms. The aim is to increase real time data processing capabilities in experimental environments such as nuclear fusion devices (e.g. ITER). The model has these features: a) real time data acquisition: the system has been provided with real time capabilities, developing specific data acquisition kernel modules under Linux and RTAI, using COMEDI project drivers; b) multiprocessor PXI (PCI extension for instrumentation) architecture: the model makes possible to add one or more processing cards (in non-system slots) to each standard PXI chassis. Several real time software modules have been developed to allow the communication between the PXI controller and the processing cards. This way the system performance is not restricted to the PXI controllers own performance. This model provides scalability to the system, adding or removing processing cards; c) real time acquired data distribution: with this model it is possible to define how to distribute, in real time, the data from all acquired signals in the system among the processing cards and the PXI controller; and d) dynamical data processing: a software platform has been developed to allow users managing dynamically their own data processing algorithms in the system. This means that users can start, stop, modify, and replace their data processing algorithms without disrupting neither the data acquisition process nor the rest of the data processing algorithms

TJ-II wave forms analysis with wavelets and support vector machines

Since the fusion plasma experiment generates hundreds of signals, it is essential to have automatic mechanisms for searching similarities and retrieving of specific data in the wave form database. Wavelet transform (WT) is a transformation that allows one to map signals to spaces of lower dimensionality. Support vector machine (SVM) is a very effective method for general purpose pattern recognition. Given a set of input vectors which belong to two different classes, the SVM maps the inputs into a high-dimensional feature space through some nonlinear mapping, where an optimal separating hyperplane is constructed. In this work, the combined use of WT and SVM is proposed for searching and retrieving similar wave forms in the TJ-II database. In a first stage, plasma signals will be preprocessed by WT to reduce their dimensionality and to extract their main features. In the next stage, and using the smoothed signals produced by the WT, SVM will be applied to show up the efficiency of the proposed method to dea...

Measurement of density and temperature fluctuations using a fast‐swept Langmuir probe

Density and temperature fluctuations have been measured in the proximity of the velocity shear location of the TJ‐I tokamak using a fast‐swept Langmuir probe technique. From the current voltage characteristic we have determined the electron temperature and the density in a time scale which is small compared with the relevant times of the turbulence. Evidence of substantial temperature fluctuations has been found near the shear location.

Adaptive high learning rate probabilistic disruption predictors from scratch for the next generation of tokamaks

The development of accurate real-time disruption predictors is a pre-requisite to any mitigation action. Present theoretical models of disruptions do not reliably cope with the disruption issues. This article deals with data-driven predictors and a review of existing machine learning techniques, from both physics and engineering points of view, is provided. All these methods need large training datasets to develop successful predictors. However, ITER or DEMO cannot wait for hundreds of disruptions to have a reliable predictor. So far, the attempts to extrapolate predictors between different tokamaks have not shown satisfactory results. In addition, it is not clear how valid this approach can be between present devices and ITER/DEMO, due to the differences in their respective scales and possibly underlying physics. Therefore, this article analyses the requirements to create adaptive predictors from scratch to learn from the data of an individual machine from the beginning of operation. A particular algorithm based on probabilistic classifiers has been developed and it has been applied to the database of the three first ITER-like wall campaigns of JET (1036 non-disruptive and 201 disruptive discharges). The predictions start from the first disruption and only 12 re-trainings have been necessary as a consequence of missing 12 disruptions only. Almost 10 000 different predictors have been developed (they differ in their features) and after the chronological analysis of the 1237 discharges, the predictors recognize 94% of all disruptions with an average warning time (AWT) of 654 ms. This percentage corresponds to the sum of tardy detections (11%), valid alarms (76%) and premature alarms (7%). The false alarm rate is 4%. If only valid alarms are considered, the AWT is 244 ms and the standard deviation is 205 ms. The average probability interval about the reliability and accuracy of all the individual predictions is 0.811 ± 0.189.

Development of an efficient real-time disruption predictor from scratch on JET and implications for ITER

Prediction of disruptions from scratch is an ITER-relevant topic. The first operations with the new ITER-like wall constitute a good opportunity to test the development of new predictors from scratch and the related methodologies. These methodologies have been based on the Advanced Predictor Of DISruptions (APODIS) architecture. APODIS is a real-time disruption predictor that is in operation in the JET real-time network. Balanced and unbalanced datasets are used to develop real-time predictors from scratch. The discharges are used in chronological order. Also, different criteria to decide when to re-train a predictor are discussed. The best results are obtained by applying a hybrid method (balanced/unbalanced datasets) for training and with the criterion of re-training after every missed alarm. The predictors are tested off-line with all the discharges (disruptive/non-disruptive) corresponding to the first three JET ITER-like wall campaigns. The results give a success rate of 93.8% and a false alarm rate of 2.8%. It should be considered that these results are obtained from models trained with no more than 42 disruptive discharges.

Automated estimation of L/H transition times at JET by combining Bayesian statistics and support vector machines

This paper describes a pattern recognition method for off-line estimation of both L/H and H/L transition times in JET. The technique is based on a combined classifier to identify the confinement regime (L or H) at any time instant during a discharge. The classifier is a combination of two different classification systems: a Bayesian classifier whose likelihood is computed by means of a non-parametric statistical classifier (Parzen window) and a support vector machine classifier. They are combined through a fuzzy aggregation operator, in particular the Einstein sum. The success rate achieved exceeds 99% for the L to H transition and 96% for the H to L transition. The estimation of transition times is accomplished by following the temporal evolution of the confinement regimes.

Search and retrieval of plasma wave forms: Structural pattern recognition approach

Databases for fusion experiments are designed to store several million wave forms. Temporal evolution signals show the same patterns under the same plasma conditions and, therefore, pattern recognition techniques can allow identification of similar plasma behaviors. Further developments in this area must be focused on four aspects: large databases, feature extraction, similarity function, and search/retrieval efficiency. This article describes an approach for pattern searching within wave forms. The technique is performed in three stages. Firstly, the signals are filtered. Secondly, signals are encoded according to a discrete set of values (code alphabet). Finally, pattern recognition is carried out via string comparisons. The definition of code alphabets enables the description of wave forms as strings, instead of representing the signals in terms of multidimensional data vectors. An alphabet of just five letters can be enough to describe any signal. In this way, signals can be stored as a sequence of ch...

Distributed real time data processing architecture for the TJ-II data acquisition system

This article describes the performance of a new model of architecture that has been developed for the TJ-II data acquisition system in order to increase its real time data processing capabilities. The current model consists of several compact PCI extension for instrumentation (PXI) standard chassis, each one with various digitizers. In this architecture, the data processing capability is restricted to the PXI controller’s own performance. The controller must share its CPU resources between the data processing and the data acquisition tasks. In the new model, distributed data processing architecture has been developed. The solution adds one or more processing cards to each PXI chassis. This way it is possible to plan how to distribute the data processing of all acquired signals among the processing cards and the available resources of the PXI controller. This model allows scalability of the system. More or less processing cards can be added based on the requirements of the system. The processing algorithms...

PXI-based architecture for real-time data acquisition and distributed dynamic data processing

This paper describes an architecture model for data acquisition systems based on compact PCI platforms. The aim is to increase real-time data processing capabilities in experimental environments such as nuclear fusion devices (e.g., ITER). The model has these features: 1) Real-time data acquisition: the system has been provided with real-time capabilities, developing specific data acquisition kernel modules under Linux and RTAI, using COMEDI project drivers; 2) Multiprocessor PCI eXtensions for Instrumentation (PXI) Architecture: the model makes possible to add one or more processing cards (in nonsystem slots) to each standard PXI chassis. Several real-time software modules have been developed to allow the communication between the PXI controller and the processing cards. This way the system performance is not restricted to the PXI controllers own performance. This model provides scalability to the system, adding or removing processing cards; 3) Real-time acquired data distribution: with this model it is possible to define how to distribute, in real-time, the data from all acquired signals in the system among the processing cards and the PXI controller; and 4) Dynamic Data Processing: a software platform has been developed to allow users to dynamically manage their own data processing algorithms in the system. This means that users can start, stop, modify, and replace their data processing algorithms without disrupting the data acquisition process or the rest of the data processing algorithms.