Aiming at the defects of traditional intrusive monitoring system in equipment input, complexity and expansibility, this paper discusses a non-intrusive load identification algorithm based on partition of the feature space. It uses K-L transform to reduce the dimension of the 10 typical features of the five typical loads, and obtains the best identification feature; It uses the least squares error algorithm to construct the discriminant function and divides the feature space of the five loads; It uses the load feature space intersection method to achieve load identification. The power consumption data acquired in the real world is used to prove that the algorithm is able to effectively achieve the load decomposition and accurately recognize the status of loads. In addition. The feature dimensionality reduction improves the efficiency of the algorithm, and the load separation improves the recognition accuracy.