Simulation and real-time replacement of missing plasma signals for disruption prediction: an implementation with APODIS

So far, the best results for real-time disruption prediction on the Joint European Torus (JET) have been achieved with the Advanced Predictor of Disruptions (APODIS). APODIS is a data-driven system whose latest version has been implemented in JET's real time-data network. It has been designed for the real-time analysis of features (mean and frequency values) corresponding to seven plasma signals in order to foresee upcoming disruptions. In this article, non-linear regression techniques are applied to create (off-line) signal models. The models are able to generate (in real-time) 'synthetic' signals. Therefore, these 'synthetic' signals can be used to replace the original ones in cases where they are in error or missing. APODIS has been tested under these conditions, emulating real-time operation. The simulation results demonstrate that once a signal in error is replaced by the generated 'synthetic' one, APODIS performance is considerably improved. The development of the regression models and the implications of the results are detailed and discussed in this paper.