Abstract: With the development of smart grid construction,the state data of distribution networks is increasing at an exponential speed. In the face of this development trend,the traditional model method based on the premise of hypothesis and simplification is no longer applicable,but the data-driven random matrix theory method shows strong advantages. In this paper,the random matrix theory,including m-p law and single ring theorem,is introduced and the methods of constructing high dimensional random matrix and the theoretical basis of using sliding time window for data sampling in the practical engineering application are summarized. The specific idea of analyzing the state of the power grid and locating the fault area is given,in which the nodal voltage amplitude is taken as the state quantity,the single loop theorem and the variance spectrum radius as the state evaluation indexes. The simulation data of the New England 10-machine 39-node network is used as a model to verify the effectiveness of the proposed method. The simulation results show that the high dimensional random matrix is suitable for the evaluation of power grid state and can effectively reflect the operation state of the power grid and locate the fault area through the single loop theorem and variance spectrum radius.