Resilient power systems urgently need real-time evaluation of their operational states. By mining the characteristics of the grid operational data and mapping them to the operational state, this paper proposes a method to evaluate the real-time state and the evolution direction of power systems. First, the state evaluation matrix is constructed using nodal voltages. Then, from the data-driven perspective, the grid state is embodied in the operational data change. Furthermore, four indicators are proposed to characterize the power grid state-from inherent physical and operating characteristics perspectives. Finally, through the simulations of a real power grid in China, it is shown that the method proposed in this paper can adequately characterize the power grid state, and is robust against bad data.