A Monitoring Model for Abnormal Electricity Consumption Based on K-Means++ Clustering and Improved K-Nearest Neighbor Algorithm

With the development of public buildings, energy consumption monitoring becomes a popular research direction. An electricity consumption abnormal data monitoring model based on K-means and K-nearest neighbor algorithm was proposed for identifying abnormal data in energy consumption monitoring. The overall architecture included two modules: detection and repair of abnormal data. Moreover, the study detected different abnormal data by partitioning these two modules separately. The initial clustering was optimized to avoid getting stuck in local optima. In addition, the slope method was introduced to improve the data repair module. These experiments confirmed that the accuracy of the research model in the Wine dataset reached 0.911. The normalized mutual information index reached 0.764. Although slightly lower than other comparison models in some datasets, the overall performance of this research model was the best. Therefore, this research model can achieve good detection and repair of abnormal electricity consumption data.