Factory Energy Data Imputation by Nearest Neighbor Search with Clustering

Accurate missing data imputation plays an important role in the analysis of energy efficiency and energy consumption of factories. For missing data imputation, a major challenge is to restore the completeness of the data while retaining the characteristics of the periodic changes of the original data to facilitate the subsequent process. This paper provides feasible solutions specifically for missing data imputation of meter-based energy data in factories. Among them, the similarity-based nearest neighbor search with the clustering imputation method can not only smoothly restore data without the problem of sudden changes but also mine the hierarchical relationship between meters and effectively restore the time-periodic features. The solutions are tested by the time series datasets from real-world meter-based scenarios. Experiments show that the solutions introduced in this paper have a greater performance improvement than traditional statistic-based and prediction-based methods.