Abstract: With the expansion of intelligent distribution network and the increasing complexity of power grid structure, the amount of data in power system increases rapidly, and new challenges rise from the checking and monitoring evaluation of power equipment. Based on the principle of big data mining analysis, this paper proposes a method based on random matrix theory and clustering algorithm to evaluate the running state of electric energy meter. Firstly, time series data of various indicators are characterized and then integrated by real-time separation window technology. Based on the random matrix theory, the random matrix-based analysis model is constructed to calculate and analyze the characteristics with multi-dimensional statistical timing in real time. Further, an improved DTW (dynamic time warping) clustering algorithm is used to analysis the linear feature statistics of the output of the random matrix. Finally, according to the clustering result, the state of the electric energy meter is obtained and outputted as different classes. The experiments show that compared with the traditional Principal Component Analysis evaluation method, the proposed method has good robustness, reliability and timeliness, which provides a new idea for the application research of power grid detection technology.

Key words: random matrix theory, clustering algorithm, moving-split window, time sequence data, power system big data, running state of electric energy meter