This repository contains the official code used in the paper Climate change drives imbalance spikes, generation droughts, and chronic seasonal synchronization in renewable power systems.
The analysis investigates how physical climate-driven shifts impact energy system extremes across multiple timescales using data-driven methods and generative learning models.
A conceptual illustration of the study framework and workflow is shown below:
The image summarizes the overall design of the data pipeline, modeling, and multi-timescale extreme energy system event evaluation.
This codebase includes the following components:
- 📈 Renewable generation and load analysis (2018–2022 hourly data)
- 🧠 Deep learning models for weather super-resolution (
train_srgan) and renewable generation predictions (train_unet) - 🧰 Common tools and data preprocessors
- 📊 Visualization templates and plotting utilities
- 💧 Hydrological impact studies in SI
| Folder | Description |
|---|---|
county/ |
County-level metadata (e.g., region mappings, FIPS codes) |
figs/ |
Figures used in the paper, including figure1.png for schematic |
future/ |
Code for future energy and climate scenario projections (2030–2050) |
hydro/ |
Hydropower-related analysis, including ecological and operational aspects |
load/ |
Load data processing: commercial, residential, industrial, and transport |
model/ |
Shared model architectures and loss functions (for SRGAN) |
plot_examples/ |
Scripts to generate paper and supplement figures |
train_srgan/ |
Super-resolution GAN (SRGAN) training scripts and configs |
train_unet/ |
U-Net model training for per capactiry renewable predicton tasks |
utils/ |
Utility functions for loading, normalizing, and evaluating data |
This codebase depends on a Google Drive dataset (~1 TB), which includes:
[Google Drive Link] (https://drive.google.com/drive/folders/1nR9cPL55tvpurUy_4ExEhLzU9bbjwgeF?usp=sharing)
- ✅ 5 full years of hourly data (2018–2022), i.e., 8760 hours/year
- ✅ Renewable generation, load, meteorology, and hydrology at high spatiotemporal resolution
- ✅ 2018 year round 8760 hour county level loads
- ✅ Future weather projections given by CMIP6 models
- ✅ AI model training checkpoints
- Described in detail in the Supplementary Information (SI) of the paper.
- Downloaded and uploaded via tools such as:
Thanks for the providers of data sources and related packages!