Publicly Available Datasets For Electric Load Forecasting An Overview

Accurate electrical load forecasting is increasingly important, particularly due to digitalization and recently emerging use cases for automated planning and control as part of the I4.0 transition in the manufacturing industry. To enable transparent research and reproducible load forecasting experiments, researchers need access to public and reusable datasets. Although there are such de-facto-standard datasets for the more general research field of time series forecasting, these are not transferable to electric load forecasting, as important external factors and industry-specific characteristics, especially in the industrial sector, are missing. This paper presents a structured literature review of existing load forecasting publications identifying suitable open-access datasets (Open Data) to help other researchers in using them. It also examines the extent to which transparent research is possible and already implemented in the field of electric load forecasting research. For this purpose, 25 unique and publicly accessible datasets were extracted from a representative set of 160 publications. The result datasets are grouped thematically, features are presented, and popularity trends are evaluated. Subsequent analysis shows a non-transparent, poorly reproducible, and methodologically weak research landscape: 54% of all publications use exclusively inaccessible private datasets for validation. Most publications (80%) use only a single dataset, 94% at most two datasets for validation - independent of journal or conference contribution. Although datasets that cover the residential and system consumption sectors are available, there are no popular public datasets showing industrial and manufacturing consumptions available for research.