The decomposed information of power consumption of household appliances is meaningful for scheduling the appliances and the reduction in home energy use. This paper validates the effectiveness of implementing co-testing in NILM. Wavelet design is used to extract features from the switching transients of loads. Using the energy coefficient of wavelet as the eigenvalue, k-NN algorithm and DT model were used to co-train the load samples. This paper has carried on the algorithmic verification experiment, has obtained the higher identification accuracy on the basis of simplifying the computational complexity and overcomes the deficiencies of OAR algorithms in the classification of true negative class events. It concludes that the research laid the foundation for the study of electricity visualization service.