For the traditional integer planning algorithms are limited to a single electrical feature quantity, resulting in poor identification of loads with similar electrical characteristics. Transient power and steady-state power are selected as load characteristics in this paper. On this basis, an improved integer programming algorithm based on low-frequency sampling is proposed. The method models the correlation between total power and various types of load power, and decomposes the total power sequence into several sub-power sequences by using the optimized segmentation algorithm. Meanwhile, the segmentation results are applied to the integer programming algorithm to improve its decomposition effect. Finally, the state switching time of each electricity load is identified on experimental data.