Ultra-LSNT

Public reproducibility repository for the manuscript:

"Controller-side feasibility guidance for wind power forecasting under corrupted SCADA"

Ultra-LSNT is the model family and experiment codebase behind the paper's controller-side forecasting study under corrupted SCADA conditions. The repository is organized for paper-aligned reproducibility rather than as a polished software package.

At a glance

• Main model: Ultra-LSNT and Ultra-LSNT-Lite
• Main task: controller-side wind power forecasting under corrupted SCADA
• Public assets: code, released Wind (CN) data assets, processed benchmark datasets, split manifests, figures, and result tables
• License: MIT

Start here

• QUICKSTART.md for the fastest local setup
• DATA.md for released datasets and split manifests
• EXPERIMENTS.md for the main experiment scripts
• STRUCTURE.md for a repository map
• docs/README.md for archived reports and maintainer notes
• CONTRIBUTING.md for contribution and issue-reporting guidance
• CITATION.cff for machine-readable citation metadata

What is in this repository

• Ultra-LSNT model implementations
• Baseline comparison scripts
• Corrupted-SCADA robustness evaluation
• Batch-1 efficiency benchmarking
• Ablation and feasibility-oriented experiments
• Cross-domain diagnostic experiments
• Public reproducibility assets referenced by the paper

Repository layout

Ultra-LSNT/
|- src/                          Models, baselines, experiments, utilities
|- scripts/                      Batch helper scripts
|- data/
|  |- raw/                       Raw Wind (CN) release
|  `- processed/                 Processed data tables
|- results/
|  |- figures/                   Paper-facing figures
|  `- tables/                    Reported and supporting result tables
|- docs/                         Archive and maintainer-facing notes
|- split_manifest_80_20.json
|- split_manifest_80_20_unified.json
`- root-level compatibility CSV files for older scripts

The canonical data location is data/. Root-level CSV copies are intentionally retained only to preserve compatibility with older scripts that expect those filenames in the repository root.

Data availability

Main Wind (CN) assets

• Raw time series: data/raw/wind_main.csv
• Processed tables:
  • data/processed/wind_final.csv
  • data/processed/processed_wind.csv
• Chronological split manifests:
  • split_manifest_80_20.json
  • split_manifest_80_20_unified.json
• Preprocessing entry point: src/data_preprocess.py

Secondary diagnostic datasets

• data/processed/wind_us.csv
• data/processed/air_quality_ready.csv
• data/processed/gefcom_ready.csv

Quick setup

git clone https://github.com/b1ue13e/Ultra-LSNT.git
cd Ultra-LSNT
pip install -r requirements.txt
python src/experiments/run_multi_domain_baselines.py --help

If the last command prints the argument list, the public repository layout is working as intended.

Main entry points

# Main Wind (CN) training
python src/experiments/train_ultra_lsnt_stable.py

# Corrupted-SCADA robustness
python src/experiments/run_universal_robustness.py

# Efficiency benchmark
python src/experiments/run_efficiency_benchmark.py

# Ablation study
python src/experiments/run_comprehensive_ablation.py

# Cross-domain diagnostics
python src/experiments/run_multi_domain_baselines.py

Reproducibility notes

• The released evaluation protocol follows chronological 80/20 splitting.
• Split definitions used in the paper are included directly in the checked-in manifest files.
• Several scripts were developed during manuscript preparation and still expect root-level filenames such as wind_final.csv; compatibility copies are retained for that reason.
• The repository is intended to support inspection and reruns of the paper's workflows, not to claim production-grade deployment readiness.

Citation

If you use this repository, the released data assets, or the corrupted-SCADA evaluation workflow, please cite:

@article{li2026ultralsnt,
  title   = {Controller-side feasibility guidance for wind power forecasting under corrupted SCADA},
  author  = {Li, Junyu and Du, Juntao},
  journal = {Applied Energy},
  year    = {2026},
  note    = {Under review}
}

License

This project is released under the MIT License. See LICENSE for details.