The recognition result of the single feature may be inconsistent with the actual electricity consumption situation, and the residential electricity behavior is one of the important factors that affect the load identification, in order to improve the decomposition ability of non- intrusive electric load more reliably, it is proposed to use residential electricity behavior as one of the characteristics of load identification, and through the structural entropy method, the residential electricity behavior is related to active power, reactive power and current harmonics feature combination of the multi-feature recognition algorithm, that is, the quantitative analysis of the entropy method andqualitative analysis of the subjective assignment method to determine the final weight, access to load recognition results. Finally, the multi-feature recognition algorithm with structural entropy method is compared with the single feature and the load recognition mode which does not take into account the behavior of residents. The results show that the multi-feature recognition algorithm considering the behavior of residents can effectively improve the accuracy of load identification.