Burn conditions stabilization with artificial neural networks of subignited thermonuclear reactors with scaling law uncertainties

In this work it is demonstrated that robust burn control in long-pulse operations of subignited thermonuclear reactors can be achieved with radial basis neural networks (RBNNs) composed of Gaussian nodes in the hidden layer and sigmoidal units in the output layer. The results reported here correspond to a volume-averaged zero-dimensional nonlinear model of a subignited fusion reactor with design parameters corresponding to those of the ITER-EDA group. The control actions are implemented through the concurrent modulation of the D-T refuelling rate, a neutral He-4 beam and an auxiliary heating power, constrained to lie below maximum allowable levels. It is shown that the resulting network provides feedback stabilization over a wide range of energy confinement times for plasma density and temperature excursions significantly far from their nominal operating values. The results show that the RBNN feedback-controlled nonlinear system is stable regardless of any particular scaling law, as long as the confinement time lies within the scope of the training region. In addition, it also shows robustness with respect to noise in the energy confinement time value fed into the controller during simulated transients of a thermonuclear system using a particular ELMy scaling law, as well as with respect to the thermalization time of the alpha particles produced by fusion.