New position control tools for runaway experiments at JET

Runaway beam confinement and dissipation remain one of the main concern for ITER operation and a clear solution has not been found yet. ITER will be provided with a Shattered Pellet Injection (SPI) system as the primary disruption mitigation technique given the promising results provided by DIII-D [3]. To further study such technique an SPI system has been recently installed at JET and to provide reliable results an improved runaway beam position control system [2, 1] is proposed. We propose to use a dynamic observer to estimate in realtime the slow vertical drift of the runaway beam. This dynamic observer should replace the static one once the runaway beam is detected. The observer parameters have been optimized in order to fit the vertical position zp reconstructed using EFIT. The new observer has the same high frequency behavior of the standard one plus the capability of detecting the RE beam slow vertical drift. An innovative tool to improve the beam position control is also described. This method uses a graph data structure to store an adaptive probabilistic route-map that links different states of the plasma and that can be obtained either using experimental data or via simulators. Such structure is then used to provide an optimal control feedforward IP4 current references via reinforcement learning techniques.