Early anomaly detection of distribution network based on large-dimensional matrix spectrum analysis

Abstract

Aiming at the complex structure of distribution network and the interaction of many factors, a method for early fault detection of distribution network based on random matrix theory is proposed. Firstly, we propose to construct a spatio-temporal data matrix by using the measurement data of all distribution transformers within distribution network feeders and their branches. Then the sliding window method and random matrix theory are used to continuously analyze the statistical properties of the elements in the data matrix. Linear eigenvalue statistics are used as statistical indicators to characterize the behavior of the data for realizing anomaly detection. The test results of the simulation data and the engineering data show that the approach can effectively realize early anomaly event detection in distribution network, and has certain guiding significance for distribution network line inspection.