A hybrid control system with a discrete adaptive predictive model for a nonstationary unstable third-order dynamic plant in continuous time is synthesized and modeled. An adaptive state observer, estimating a variable parameter of the plant model with respect to the a quadratic quality criterion, was synthesized. Continuous estimation of the plant parameter for a discrete sample is used in a discrete adaptive control algorithm with a predictive model. A linear model of the control plant mimics the unstable vertical motion of plasma in a tokamak with a vertical cross-section elongated along the vertical axis compared to a given equilibrium position.