G-DDPM: Graph-based Denoising Diffusion Probabilistic Model

Probabilistic Forecasting of Renewable Energy and Electricity Demand

This repository contains the official implementation of the paper:

Probabilistic Forecasting of Renewable Energy and Electricity Demand using Graph-based Denoising Diffusion Probabilistic Model (G-DDPM)
Amir Miraki, Pekka Parviainen, Reza Arghandeh
Published in Energy and AI, Volume 19, 2025

🌟 Highlights

• 🔄 Diffusion-based probabilistic forecasting for multivariate time series
• 🌐 Graph-based modeling of variable interdependencies
• ⏱️ Parallel feature extraction for both spatial and temporal information
• 📊 State-of-the-art performance on real-world energy datasets
• 🧠 Incorporates GRU, Gated Graph Convolutions, and Dilated Convolutions
• 📉 Outperforms DeepAR, TimeGrad, GCRDD, SpecSTG, and others on MAE, RMSE, CRPS

📖 Paper Abstract

The paper introduces G-DDPM, a novel diffusion-based generative model that combines temporal modeling with graph-based representations to capture dependencies between variables in multivariate time series. Tested on two real-world datasets (renewable energy and electricity demand), G-DDPM significantly outperforms existing forecasting models in terms of accuracy and uncertainty quantification.

🔗 Full paper: ScienceDirect

📁 Datasets

This repo uses two public datasets:

1. Renewable Energy: CAISO & NREL data (solar, wind, weather, demand)
2. Electricity Demand: Hourly load of 370 clients from a public benchmark

📈 Performance Summary

Method	MAE ↓	RMSE ↓	CRPS ↓
DeepAR	High	High	High
SpecSTG	Medium	Medium	Medium
TimeGrad	Good	Good	Good
G-DDPM	Best	Best	Best

🚀 Getting Started

Requirements

• Python 3.8+
• PyTorch
• NumPy
• NetworkX
• tqdm

Installation

git clone https://github.com/AmirMiraki/GDDPM.git
cd GDDPM