An Imputation Method Based on Dummy Variable and Unsupervised Learning for Electricity Consumption Data with Missing Values

Due to the fault of smart meters, system maintenance, data storage, and other reasons, part of the electricity consumption data is missing, which brings some difficulties to the behavior analysis of electricity consumption data. And deleting the missing samples will cause a loss of electricity information and a lower accuracy of detection model. Aiming at these problems, this paper proposes an imputation method based on dummy variable and unsupervised learning, and impute the missing values of electricity consumption data without affecting the quality of data. First, the dummy variable considers the missing information mode in the users’ electricity consumption dataset and saves the important missing information. Combined with the deep learning model to learn the potential feature representation of electricity consumption dataset, and effectively realizes the complex relationship between variables through the nonlinear transformation. The generated network and discriminant network of the unsupervised learning are employed to generate the missing values and discriminate the imputation values to reduce the model error, thus providing a great imputation model to estimate the missing values of electricity consumption dataset. The root mean square error (RMSE) of different imputation models based on different datasets and different missing rates verifies that the proposed missing imputation model can more accurately and efficiently impute the missing values of electricity consumption data.