Probabilistic Imputation for High Resolution Univariate Electric Load Data with Large Gaps

Abstract

The use of electric load data within power engineering applications is critical. Such data often contain missing observations, especially higher resolution datasets important for detailed modeling and simulation. Missing data frequently are handled by replacement with new values, in other words imputation. However, most readily available imputation methods will perform unsatisfactorily on electric load data when many successive observations are missing as they cannot capture the periodic variation. As well many methods only provide single point estimates, allowing for no assessment of probabilistic characteristics of the missing data. In this study a new imputation method is proposed that captures the periodic variation common in high resolution load data, as well as generate probabilistic and point estimates for the missing data. The method is evaluated on three real-world high resolution load datasets and compared with a typical imputation technique.