Investigation of plasma dynamics to detect the approach to the disruption boundaries

This contribution shows a machine learning technique to detect when the plasma starts to move away from a safe region in the parameter space and, therefore, it begins to approach a disruptive zone. The recognition method is based on an accurate and reliable disruption predictor based on the formalism of conformal predictions. The technique allows following the plasma dynamics during a discharge. In general, the plasma evolves in a steady way, far enough from the disruptive region. However, some events can push the plasma towards the dangerous region. Most of times, the plasma goes back to the initial safe zone. But sometimes, the plasma reaches a non-return point and the disruption is inevitable. This has been analysed with 297 JET unintentional disruptions during the ILW campaigns. On average, the transit from the non-return point and the disruption takes 749 ms (with a standard deviation of 1.060 s). This transit shows three phases. In a first period, the plasma goes from the non-return point (in the safe region) to an intermediate zone, where the plasma state (disruptive or non-disruptive) is not clearly defined. On average, this first phase takes 180±436 ms. In the second phase, the plasma maintains its undefined state during 244±686 ms. Finally, the plasma enters the disruptive region and this phase lasts 325±536 ms.