Smart meters are the core device for electricity metering and billing, which is related to the economic interests of owners, power grids and other parties. It has the characteristics of a large number and complex operating environment. In order to effectively carry out online operation and inspection of massive smart meters, a random matrix-based method for locating abnormal individuals of massive smart meters is proposed. First, a number of parameter characterization methods for the health of smart meters are proposed, including non-electrical parameters such as ratio difference, angular difference, temperature, humidity, vibration, and electrical parameters such as primary voltage and magnetic field. Secondly, in order to evaluate the state of the smart meter more accurately and comprehensively, the data obtained from the test of the smart meter, simulation data and historical operating data are used as the data source, and the parameters of the health state of the smart meter are selected to construct a high-dimensional random matrix and analyze it, to realize the positioning of abnormal individuals of smart meters. Finally, the actual data of the new generation smart meter in China Southern Power Grid is used to verify the effectiveness of the method in this paper.