The invention relates to a method of constructing a random matrix for distribution network reactive power optimization, comprising the following steps: S1, acquiring a random matrix original data source containing historical load data; S2, extracting load data, and constructing a load random matrix data set; S3, getting a cumulative load function according to the load random matrix data set, and constructing a load random matrix; S4, getting a covariance matrix through the load random matrix; S5, getting the average spectral radius based on the ring law; and S6, getting a reactive power optimization control sequence through a distribution network reactive power optimization big data method according to the average spectral radius. According to the technical scheme of the invention, the power load is expressed in the form of a random matrix, and both the temporal distribution characteristics and the spatial distribution characteristics of the power load in a distribution network are considered. The method can reflect the distribution characteristics of load spectrum quantitatively, and can be used in load distribution characteristic comparing and matching as well as distribution network reactive power optimization under a framework of big data.