This study explores the development of a data-driven control strategy for complex nonlinear systems. Based on the stream of temporal measurements, the online dynamic mode decomposition discovers the underlying nonlinear system and transforms it to a linear time-varying realization. The model predictive control is then designed for the discovered linear time-varying model. To address the problem of potential fluctuations in the extracted model that could lead to system divergence, tracking error is utilized as a criterion to determine whether a new model should be obtained. The efficacy of the proposed approach in governing complex nonlinear systems, such as the chaotic Lorenz system, is demonstrated. Simulation studies illustrate the high efficiency of the proposed method in both the stabilization and regulation control of the Lorenz and Van der Pol systems.
