Burn stabilization of a tokamak power plant with on-line estimations of energy and particles confinement times

Using a two-temperature volume averaged 0D model we show that robust stabilization of the burn conditions can be achieved by means of Radial Basis Neural Networks (RBNN) in a tokamak power plant in which transport properties of the plasma system, represented by the energy and helium ash confinement times, are only known through on-line estimations. In the model we take into account alpha particle thermalization time delay and different amount of energy deposited to the electrons and to the ions by the α's during thermalization. The control actions include the modulation of the DT refueling rate, a neutral He-4 injection beam and auxiliary heating powers to ions and electrons constrained to lie within allowable range values. To demonstrate the robustness of this controller transients with on-line estimations of both the energy and the helium ash confinement time are simulated assuming the system follows the IPB98(y,2) scaling for ΤE and an arbitrary independent scaling for Τα; we simulate noisy measurements of these two quantities by corrupting them with pseudo Gaussian noise before been input into the RBNN. (author)