WPBench: A Comprehensive Benchmark for Wind Power Forecasting

WPBench benchmarks 19 forecasting models on 26 public wind-power datasets. This repository provides fixed-parameter scripts for the paper's main results and foundation-model adaptation experiments.

Quickstart

Run the following commands from the repository root.

1. Environment

conda env create -n wpbench -f environment/wpbench_unified_hpo.yml
conda activate wpbench

2. Datasets

Download the 26-dataset bundle and restore the CSV files and metadata:

mkdir -p dataset/forecasting/forecasting_wp1_all_shapes_v1
cp -a /path/to/wpbench_26_bundle_20260611/forecasting_wpbench_26/*.csv \
  dataset/forecasting/forecasting_wp1_all_shapes_v1/
cp /path/to/wpbench_26_bundle_20260611/metadata/FORECAST_META.csv \
  dataset/forecasting/FORECAST_META.csv

Keep the supplied CSV filenames and format (date,data,cols). To load data from another location, set WPBENCH_FORECASTING_DATASET_PATH to the directory containing FORECAST_META.csv and forecasting_wp1_all_shapes_v1/.

3. Model weights

For FactoST_STA, FactoST_UTP, OpenCity_STFM, SEMPO, TinyTimeMixer, and Toto, download the checkpoint bundle and restore it before running their scripts:

mkdir -p checkpoints
cp -a /path/to/wpbench_6algos_checkpoints_20260611/checkpoints/. checkpoints/

4. Run an experiment

All experiment scripts are in scripts/run_experiments/final_results/. For example, run DLinear on Yalova:

bash scripts/run_experiments/final_results/DLinear/standard/tfb-Yalova_final_ready/horizon_12/run.sh