An efficient abnormal load identification method in large-scale power system employing Rayleigh quotient and parallel computing technology is proposed for improving the efficiency of abnormal load identification and the adaptability in large-scale power system. It firstly constructs a group of data source matrices with random load fluctuation and noise interference for each area, respectively. Then their standard matrices would be obtained through using a moving window matrix. Consequently, the sample covariance matrices of area can be formed. Secondly, the MESCM of area could be estimated rapidly with the help of the Rayleigh quotient and the parallel computing technology. In this way, the rapid identification of abnormal load in large-scale power system may be achieved by the violation checking of MESCM. The case studies have been carried on an IEEE 54-machine and 118-bus system and a Polish 420-machine and 2736-bus system utilizing Matlab R2014a and PST software. The results in comparison with those results from the traditional mean spectral radius based method show that the proposed methodology is valid and efficient. This work is supported by National Natural Science Foundation of China (No. 51567006), Program for Top Science＆Technology Talents in Universities of Guizhou Province (No. 2018036), and Science and Technology Fund of Guizhou Province (No. [2019]1100).