We report on the burn control studies of a D-T-fueled tokamak reactor using a two-temperature, zero-dimensional, volume-averaged model, assuming that electrons and ions have the same radial profile with different central temperatures. Balance equations for the particle and energy densities are used assuming that energy and particle transport losses are independent of each other and can be estimated online; thermalization time delays of the energetic alpha particles produced by fusion are taken into account in the dynamical equations. The burn stabilization is achieved with radial basis neural networks (RBNNs) that concurrently modulate a D-T refueling rate, a neutral 4He beam, and auxiliary heating powers to the electrons and the ions, all constrained to maximum allowable levels. The resulting network provides feedback stabilization in a wide range of energy confinement times for plasma density and temperature excursions significantly far from their nominal values. Transient examples using different ELMy scaling laws show that the RBNN controller is stable with respect to any particular scaling law that the tokamak may actually follow for the energy and particle transport losses and is also robust with respect to noise in the measurement of the confinement times. Furthermore, it satisfactorily responds to sudden changes in fast-alpha-particle losses due to increments in magnetohydrodynamic events.