This paper presents a new adaptive controller via multilevel technique. The control strategy is based on decomposing control signals into two components: the first one balances the effect of interconnections between the subsystems, while the second one is a local adaptive controller based on hyperstability theory for the decentralized subsystems. The changeable gains generated by a non-linear time-varying function act as the adaptive mechanism. Compared with MRAC or STAC the proposed control avoids the difficulty in choosing an appropriate reference model and the burden of implementing an online parameter estimator. The local control is constructed so that the closed-loop system is hyperstable guaranteeing the dynamic stability improvement of each subsystem in the multimachine power system. A guarantee of stability of the overall system is also given. Simulation results of a numerical example of a multimachine power system reveal that the low frequency oscillations can completely be eliminated using the proposed control scheme.