In the context of non-intrusive load monitoring for residential users, there exist similar appliances with comparable power consumption and internal circuitry. However, the existing algorithms have a low success rate in identifying these appliances. In order to improve the accuracy, this paper proposes a non-intrusive appliance identification algorithm based on multivariate feature analysis. The method uses a pair of multidimensional low-frequency appliance feature data for analysis, which are first standardized, and then the Mahalanobis distance between the two appliance features is calculated to determine their similarity. The original features are then subjected to principal component analysis to extract the main features of this pair of similar appliances. These features are then input into a multivariate Gaussian model to obtain the identification results. This can judge the operating status of the appliance and measure its energy consumption. The proposed algorithm is validated using actual appliance data and residential electricity consumption data, and compared with other models. The results demonstrate that the algorithm can effectively improve the accuracy of identifying similar appliances.