EnergyDiff

The public repository of diffusion-based energy time series data generation. Paper: EnergyDiff: Universal Time-Series Energy Data Generation using Diffusion Models.

Installation

1. Install the required dependency packages. This is the easiest to be done with uv. Search for and install the package manager uv first. Then,

cd path/to/this/repository
uv sync

Run

1. Prepare the datasets by downloading from their original sources and running the preprocessing scripts (under scripts/python/preprocessing).

2. Training, sampling, and evaluation. Refer to the scripts (both for bash, for slurm, and for python) under scripts/ directory.

Licence

This repository adopts an MIT License. The conditional copula model is implemented by conditonal-copula .