The invention relates to a regional power grid oscillation online early-warning method and a regional power grid oscillation online early-warning device. The regional power grid oscillation online early-warning method comprises the steps of: acquiring current power flow data, and performing small interference calculation on the current power flow data to obtain an intrinsic oscillation mode of a system and eigenvalues corresponding to the intrinsic oscillation mode; acquiring oscillation fluctuation signals injected by a power grid node within historical time, clustering the oscillation fluctuation signals injected by the power grid node according to eigenvalues corresponding to the oscillation fluctuation signals injected by the power grid node, and acquiring a clustering center of the oscillation fluctuation signals injected by the power grid node and eigenvalues corresponding to the clustering center; and performing power grid online early-warning according to similarity between theeigenvalues corresponding to the clustering center and the eigenvalues corresponding to the intrinsic oscillation mode, wherein the eigenvalues comprise oscillation frequency and a damping ratio. Theregional power grid oscillation online early-warning method and the regional power grid oscillation online early-warning device provided by the invention improve the insufficiency of previous regional power grid oscillation online early-warning methods in various aspects, and improve the recognition quality of regional oscillation modes of a unit abnormal fluctuation trigger unit.