Quick Links Website • Paper • Hugging Face • Leaderboard • Energy-TTM • Energy-TSPulse
EnergyFM is a suite of pretrained time series foundation models (TSFMs) for energy meter data analytics. It provides pretrained models, example pipelines, and tutorial notebooks for short-term load forecasting, anomaly detection, and appliance classification on large-scale smart meter datasets.
EnergyFM is designed to support:
- Zero-shot inference on unseen buildings and regions
- Fine-tuning for downstream energy analytics tasks
- Reproducible benchmarking across forecasting and representation-learning tasks
This repository accompanies the paper:
EnergyFM: Pretrained Models for Energy Meter Data Analytics ACM e-Energy 2026
Important
EnergyFM has moved.
This project is now maintained under the EnergyFM GitHub organization. For the latest code, model releases, documentation, and updates, please visit:
github.com/energyfms
EnergyFM currently includes two pretrained models, each targeting a different class of energy analytics tasks:
| Model | Architecture | Primary Use | Supported Tasks |
|---|---|---|---|
| Energy-TTM | Tiny Time Mixer (TTM) | Forecasting | Short-term load forecasting |
| Energy-TSPulse | TSPulse | Representation learning | Anomaly detection, appliance classification |
To compare EnergyFM against other state-of-the-art time series foundation models for energy analytics, visit our public benchmark leaderboard:
👉 Energy Benchmark Leaderboard
The leaderboard provides standardized evaluations across multiple energy forecasting, anomaly detection, and classification tasks, allowing direct comparison between EnergyFM and other TSFMs under consistent evaluation settings.
-
Zero-Shot Forecasting Open Notebook
-
Fine-Tuning for Forecasting Open Notebook
-
Zero-Shot Anomaly Detection Open Notebook
-
Fine-Tuning for Anomaly Detection Open Notebook
-
Appliance Classification (Fine-Tuning) Open Notebook
Pretrained checkpoints and model cards are hosted on Hugging Face:
-
Energy-TTM — forecasting model for short-term load forecasting View Model Card
-
Energy-TSPulse — representation model for anomaly detection and appliance classification View Model Card
EnergyFM builds on IBM’s Granite Time Series Foundation Models (TSFM) ecosystem, which provides the core model implementations and Hugging Face integration.
- Granite TSFM GitHub: ibm-granite/granite-tsfm
- Granite TSFM Wiki: Granite TSFM Wiki
If you use EnergyFM in your research, please cite:
@inproceedings{energyfm,
author = {Arjunan, Pandarasamy and Srivastava, Naman and Kumar, Kajeeth and Jati, Arindam and Ekambaram, Vijay and Dayama, Pankaj},
title = {EnergyFM: Pretrained Models for Energy Meter Data Analytics},
year = {2026},
isbn = {9798400720116},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3744255.3798119},
doi = {10.1145/3744255.3798119},
booktitle = {Proceedings of the 17th ACM International Conference on Future and Sustainable Energy Systems},
pages = {556--568},
series = {E-Energy '26}
}
Please report bugs, feature requests, or questions via GitHub Issues: