Abstract: With the advancement of intelligent power systems, the integrity of electricity consumption data imposes heightened demands on load forecasting and dispatch. Addressing the limitations of the traditional K-means algorithm, such as sensitivity to initial clustering centers, susceptibility to local optima,and the issue of missing electricity data, this study proposes a combined repair method integrating an enhanced clustering algorithm with compressed sensing, setting up data missing scenarios with low attrition rate, high attrition rate, and continuous missing rates for experimental verification. Then the particle swarm optimization (PSO) algorithm is employed to implement global optimal clustering center search, utilizing the silhouette coefficient and CH index to verify the clustering performance of the PSO-K-means algorithm; based on the clustering results of electricity consumption data obtained using the PSO-K-means algorithm, pre-fill missing time periods with the mean value of similar data, and perform compressive sensing reconstruction on the time series constructed from similar data. Results demonstrate that among the three scenarios set, compared with other methods, the proposed method excels in both the coefficient of determination and root mean square error indicators, significantly enhancing the accuracy of data repair. It provides an innovative technical path for optimizing data quality in smart grids and effectivelys upports precise scheduling and operation of power systems.

    Key words: PSO-K-means algorithm; compressed sensing; electricity consumption data; data repair