LM-Load-Forecasting

Introduction

This folder includes the code and more details of our paper: Utilizing Language Models for Energy Load Forecasting, which will be presented at BuildSys 2023. This work is developed from our previous work: PromptCast: A New Prompt-based Learning Paradigm for Time Series Forecasting

Its code is also available here.

How to USE

1. prompt your data (describe your energy usage csv data to language sentences): use data_prompting.py
2. prepare train/val/test set data: use prepare_hf.py
3. fine-tune language models: use run_hf_s2s.py
4. test your fine-tuned language models with your test set: user run_inference.py

• For step 3 and 4, there are examples provided in example.sh
• Our PromptCast Repo also provides more details, you can check here.

Note

If you think our paper/code is useful, please cite our paper:

@inproceedings{10.1145/3600100.3623730,
author = {Xue, Hao and Salim, Flora D.},
title = {Utilizing Language Models for Energy Load Forecasting},
year = {2023},
publisher = {Association for Computing Machinery},
booktitle = {Proceedings of the 10th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation},
pages = {224–227},
numpages = {4},
location = {Istanbul, Turkey},
series = {BuildSys '23}
}

The dataset used in our paper is private data, and we can't share them for now. If we get the green light, I'll modify this repo further