StreamFuels: A Dynamic Resource for Time Series Benchmarking in Fuel Analytics

This repository contains the datasets obtained from StreamFuels on April 2026. StreamFuels provides access to five curated datasets comprising thousands of time series representing fuel sales and industrial operations across all 27 Brazilian states and over 5,500 cities. StreamFuels automatically retrieves the most recent data from official sources at access time, ensuring users always work with clean, current data. This dynamic capability supports benchmarking under realistic, time-evolving scenarios and enables the study of model robustness and adaptation over time. StreamFuels is publicly available via the Python Package Index at https://pypi.org/project/streamfuels/. The source code and results are also available at https://github.com/lucas-castrow/StreamFuels. A summary of the datasets is provided in the following Yearly Fuel Sales by State: 216 series of yearly sales for eight fuel types across 27 states. The earliest records date back to 1947. Yearly Fuel Sales by City: 29,464 series of yearly sales for eight fuels and asphalt across 5,325 cities. Monthly Fuel Sales by State: 216 series of monthly sales for eight fuel types across 27 states, all dating back to 1990. Monthly Oil and Gas Operations by State: 76 series of monthly records for five types of industrial operations (production, reinjection, flaring, self-consumption, and availability) across three fuel types (natural gas, petroleum, and NGL). Fuel Type Classification: 14,409 fixed-length series (12 observations each) with eight class labels; most classes are balanced, except for kerosene-i and fuel oil. All datasets are in Time Series Format (TSF), compatible with most popular time series toolkits, such as aeon and sktime. Please, use the following reference when citing these datasets: @inproceedings{streamfuels, title={StreamFuels: A Dynamic Resource for Time Series Benchmarking in Fuel Analytics}, author={Castro, L. G. M. and Ribeiro, A. G. R. and Barddal, J. P. and Britto Jr, A. S. and Souza, V. M. A.}, booktitle={Conference on Information and Knowledge Management (CIKM)}, pages={1--4}, year={2026}, organization={ACM}, note={https://doi.org/10.5281/zenodo.19444005}}