Fusion Engineering and Design 96–97 (2015) 698–702 Contents lists available at ScienceDirect Fusion Engineering and Design jo u r n al homep age: www.elsevier.com/locate/fusengdes Improvements in disruption prediction at ASDEX Upgrade R. Aledda, B. Cannas, A. Fanni, A. Pau, G. Sias ∗ , the ASDEX Upgrade Team Department of Electrical and Electronic Engineering, University of Cagliari, Italy h i g h l i g h t s • A disruption prediction system for AUG, based on a logistic model, is designed. • The length of the disruptive phase is set for each disruption in the training set. • The model is tested on dataset different from that used during the training phase. • The generalization capability and the aging of the model have been tested. • The predictor performance is compared with the locked mode detector. a r t i c l e i n f o Article history: Received 29 September 2014 Received in revised form 19 March 2015 Accepted 23 March 2015 Available online 16 April 2015 Keywords: Disruption prediction Nuclear fusion Logistic regressor Mahalanobis distance a b s t r a c t In large-scale tokamaks disruptions have the potential to create serious damage to the facility. Hence disruptions must be avoided, but, when a disruption is unavoidable, minimizing its severity is mandatory. A reliable detection of a disruptive event is required to trigger proper mitigation actions. To this purpose machine learning methods have been widely studied to design disruption prediction systems at ASDEX Upgrade. The training phase of the proposed approaches is based on the availability of disrupted and non-disrupted discharges. In literature disruptive configurations were assumed appearing into the last 45 ms of each disruption. Even if the achieved results in terms of correct predictions were good, it has to be highlighted that the choice of such a fixed temporal window might have limited the prediction performance. In fact, it generates confusing information in cases of disruptions with disruptive phase different from 45 ms. The assessment of a specific disruptive phase for each disruptive discharge represents a relevant issue in understanding the disruptive events. In this paper, the Mahalanobis distance is applied to define a specific disruptive phase for each disruption, and a logistic regressor has been trained as disruption predictor. The results show that enhancements on the achieved performance on disruption prediction are possible by defining a specific disruptive phase for each disruption. © 2015 Elsevier B.V. All rights reserved. 1. Introduction The physical phenomena leading to plasma disruptions in toka- maks are very complex and non-linear, and the present state of knowledge is not sufficient to explain the intrinsic structure of the data of interest. Following a data based approach, the plasma in a tokamak is considered as a multidimensional system which can be described through a multidimensional set of observables, i.e., available diagnostic signals recorded during the experiments. ∗ Corresponding author. Tel.: +39 0706755878; fax: +39 0706755900. E-mail addresses: raffaele.aledda@diee.unica.it (R. Aledda), cannas@diee.unica.it (B. Cannas), fanni@diee.unica.it (A. Fanni), alessandro.pau@diee.unica.it (A. Pau), giuliana.sias@diee.unica.it (G. Sias). To this purpose data-based methods have been widely studied to design disruption prediction systems on several experimental devices. In particular, for ASDEX Upgrade (AUG), literature pro- posed predictive systems applying data-based techniques, such as Multi-layer Perceptron neural networks [1,2], Discriminant Analy- sis [3], and Self-Organizing Maps [4]. Moreover, for JET the neural network approaches have been widely applied for disruption pre- diction both for carbon wall [5–7] and full metal ITER-like wall [8]. For the spherical torus NSTX a disruption predictor based on a multivariate statistical analysis of diagnostic data is proposed [9]. The training phase of the proposed approaches is based on the availability of disrupted and non-disrupted discharges. To accom- plish an exhaustive model, every disruptive and safe configuration included in the machine operational space should be represented in the training set. Safe configurations were selected from safe http://dx.doi.org/10.1016/j.fusengdes.2015.03.045 0920-3796/© 2015 Elsevier B.V. All rights reserved.