In the process of power spot market settlement, it is very important to obtain the real-time power of market users. This paper focuses on the power decomposition of non-time-sharing metering users in the spot market and designs a method and process to obtain time-sharing power using typical load curve. Sample users are selected, and the complete sample typical load curve is obtained after preprocessing the measurement data of sample users. Then, this paper proposes a clustering center based on kernel density estimate the load curve of clustering method, the kmeans algorithm the original average clustering center to get upgraded to a Gaussian kernel density estimation to obtain the maximum probability of iterative calculation, the clustering center and clustering center curve as a typical load curve of users that do not have time-sharing measurement, power division, division to the granularity for 15 min settlement of electricity, using the sample user measurement data of Yunnan province, the traditional peak pinggu scale decomposition, traditional clustering algorithm and the improved clustering algorithm to obtain the typical load curve of power of the real example analysis, the results show that the improved kmeans algorithm proposed in this paper has better classification performance and better efficiency, at the same time decomposition capacity have higher accuracy.