AI4Climate/WindFarm_raw

Background

The SDWPF dataset, collected over two years from a wind farm with 134 turbines, details the spatial layout of the turbines and dynamic context factors for each. This dataset was utilized to launch the ACM KDD Cup 2022, attracting registrations from over 2,400 teams worldwide. To facilitate its use, we have released the dataset in two parts: sdwpf_kddcup and sdwpf_full. The sdwpf_kddcup is the original dataset used for the Baidu KDD Cup 2022, comprising both training and test datasets. The sdwpf_full offers a more comprehensive collection, including additional data not available during the KDD Cup, such as weather conditions, dates, and elevation.

sdwpf_kddcup

The sdwpf_kddcup dataset is the original dataset used for Baidu KDD Cup 2022 Challenge. The folder structure of sdwpf_kddcup is:

sdwpf_kddcup
 --- sdwpf_245days_v1.csv
 --- sdwpf_baidukddcup2022_turb_location.csv
 --- final_phase_test
    --- infile
        --- 0001in.csv
        --- 0002in.csv
        --- ...
    --- outfile
        --- 0001out.csv
        --- 0002out.csv
        --- ...

The descriptions of each sub-folder in the sdwpf_kddcup dataset are as follows:

1. sdwpf_245days_v1.csv: This dataset, released for the KDD Cup 2022 challenge, includes data spanning 245 days.

2. sdwpf_baidukddcup2022_turb_location.csv: This file provides the relative positions of all wind turbines within the dataset.

3. final_phase_test: This dataset serves as the test data for the final phase of the Baidu KDD Cup. It allows for a comparison of methodologies against those of the award-winning teams from KDD Cup 2022. It includes an 'infile' folder containing input data for the model, and an 'outfile' folder which holds the ground truth for the corresponding output. In other words, for a model function y = f(x), x represents the files in the 'infile' folder, and the ground truth of y corresponds to files in the 'outfile' folder, such as {001out} = f({001in}).

More information about the sdwpf_kddcup used for Baidu KDD Cup 2022 can be found:

@article{zhou2022sdwpf,title={SDWPF: A Dataset for Spatial Dynamic Wind Power Forecasting Challenge at KDD Cup 2022}, author={Zhou, Jingbo and Lu, Xinjiang and Xiao, Yixiong and Su, Jiantao and Lyu, Junfu and Ma, Yanjun and Dou, Dejing}, journal={arXiv preprint arXiv:2208.04360},year={2022}}

sdwpf_full

The sdwpf_full dataset offers more information than what was released for the KDD Cup 2022. It includes not only SCADA data but also weather data such as relative humidity, wind speed, and wind direction, sourced from the Fifth Generation of the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalyses of the global climate (ERA5). The dataset encompasses data collected over two years from a wind farm with 134 wind turbines, covering the period from January 2020 to December 2021. The folder structure of sdwpf_full is:

sdwpf_full
--- sdwpf_turb_location_elevation.csv
--- sdwpf_2001_2112_full.csv
--- sdwpf_2001_2112_full.parquet

The descriptions of each sub-folder in the sdwpf_full dataset are as follows:

1. sdwpf_turb_location_elevation.csv: This file details the relative positions and elevations of all wind turbines within the dataset.

2. sdwpf_2001_2112_full.csv: This dataset includes data collected two years from a wind farm containing 134 wind turbines, spanning from Jan. 2020 to Dec. 2021. It offers comprehensive enhancements over the sdwpf_kddcup/sdwpf_245days_v1.csv, including:

   • Extended time span: It spans two years, from January 2020 to December 2021, whereas sdwpf_245days_v1.csv covers only 245 days.

   • Enriched weather information: This includes additional data such as relative humidity, wind speed, and wind direction, sourced from the Fifth generation of the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalyses of the global climate (ERA5).

   • Expanded temporal details: Unlike during the KDD Cup Challenge where timestamp information was withheld to prevent data linkage, this version includes specific timestamps for each data point.

3. sdwpf_2001_2112_full.parquet: This dataset is identical to sdwpf_2001_2112_full.csv, but in a different data format.