SDWPF: A Dataset for Spatial Dynamic Wind Power Forecasting over a Large Turbine Array

<br><b>Paper</b>This dataset is associated with the paper published in Scientific Data, titled "SDWPF: A Dataset for Spatial Dynamic Wind Power Forecasting over a Large Turbine Array." You can access the paper: <i>https://www.nature.com/articles/s41597-024-03427-5</i>If you find this dataset useful, please consider citing our paper: <b>Scientific Data Paper</b>@article{zhou2024sdwpf, title={SDWPF: A Dataset for Spatial Dynamic Wind Power Forecasting over a Large Turbine Array}, author={Zhou, Jingbo and Lu, Xinjiang and Xiao, Yixiong and Tang, Jian and Su, Jiantao and Li, Yu, and Liu, Ji and Lyu, Junfu and Ma, Yanjun and Dou, Dejing},journal={Scientific Data},volume={11},number={1},pages={649},year={2024},url = {https://doi.org/10.1038/s41597-024-03427-5},publisher={Nature Publishing Group}}<br><b>Baidu KDD Cup Paper</b>@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},url = {https://arxiv.org/abs/2208.04360}, year={2022}}<br><b>Background</b>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.<b>sdwpf_kddcup</b>The <b><i>sdwpf_kddcup</i></b> dataset is the original dataset used for Baidu KDD Cup 2022 Challenge. The folder structure of sdwpf_kddcup is:<pre><pre>sdwpf_kddcup<br> --- sdwpf_245days_v1.csv<br> --- sdwpf_baidukddcup2022_turb_location.csv<br> --- final_phase_test<br> --- infile<br> --- 0001in.csv<br> --- 0002in.csv<br> --- ...<br> --- outfile<br> --- 0001out.csv<br> --- 0002out.csv<br> --- ...<br></pre></pre>The descriptions of each sub-folder in the sdwpf_kddcup dataset are as follows:<b><i>sdwpf_245days_v1.csv</i></b>: This dataset, released for the KDD Cup 2022 challenge, includes data spanning 245 days.<b><i>sdwpf_baidukddcup2022_turb_location.csv</i></b>: This file provides the relative positions of all wind turbines within the dataset.<b><i>final_phase_test</i></b>: 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 <b><i>{001out} = f({001in})</i></b>.More information about the sdwpf_kddcup used for Baidu KDD Cup 2022 can be found in Baidu KDD Cup Paper: <i>SDWPF: A Dataset for Spatial Dynamic Wind Power Forecasting Challenge at KDD Cup 2022</i><br><b>sdwpf_full</b>The <b><i>sdwpf_full</i></b> 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:<pre><pre>sdwpf_full<br>--- sdwpf_turb_location_elevation.csv<br>--- sdwpf_2001_2112_full.csv<br>--- sdwpf_2001_2112_full.parquet<br></pre></pre>The descriptions of each sub-folder in the sdwpf_full dataset are as follows:<b><i>sdwpf_turb_location_elevation.csv</i></b>: This file details the relative positions and elevations of all wind turbines within the dataset.<b><i>sdwpf_2001_2112_full.csv</i></b>: 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.<b><i>sdwpf_2001_2112_full.parquet</i></b>: This dataset is identical to sdwpf_2001_2112_full.csv, but in a different data format.<br>

**论文** 本数据集关联发表于《科学数据》（*Scientific Data*）的论文，标题为《SDWPF：面向大型风电机组阵列的空间动态风电功率预测数据集》。您可通过以下链接获取该论文：<i>https://www.nature.com/articles/s41597-024-03427-5</i>。若本数据集对您的研究有所帮助，请引用以下论文：
**《科学数据》期刊论文**
@article{zhou2024sdwpf, title={"SDWPF: A Dataset for Spatial Dynamic Wind Power Forecasting over a Large Turbine Array"}, author={Zhou, Jingbo and Lu, Xinjiang and Xiao, Yixiong and Tang, Jian and Su, Jiantao and Li, Yu, and Liu, Ji and Lyu, Junfu and Ma, Yanjun and Dou, Dejing},journal={Scientific Data},volume={11},number={1},pages={649},year={2024},url = {"https://doi.org/10.1038/s41597-024-03427-5"},publisher={Nature Publishing Group}}
**百度KDD Cup论文**
@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},url = {"https://arxiv.org/abs/2208.04360"}, year={2022}}
**背景**
SDWPF数据集采集自一座包含134台风电机组的风电场，历时两年，详细记录了风电机组的空间布局与各机组的动态关联因素。该数据集曾用于发起ACM KDD Cup 2022赛事，吸引了全球超过2400支队伍报名参赛。为便于使用，我们将数据集分为两个版本发布：sdwpf_kddcup与sdwpf_full。其中sdwpf_kddcup为2022年百度KDD Cup的原始数据集，包含训练集与测试集；sdwpf_full则为更全面的数据集，包含了KDD Cup赛事期间未公开的额外数据，例如气象条件、日期信息与海拔高度。
**sdwpf_kddcup**
sdwpf_kddcup数据集为2022年百度KDD Cup挑战赛的原始数据集，其目录结构如下：
<pre><pre>sdwpf_kddcup
--- sdwpf_245days_v1.csv
--- sdwpf_baidukddcup2022_turb_location.csv
--- final_phase_test
--- infile
--- 0001in.csv
--- 0002in.csv
--- ...
--- outfile
--- 0001out.csv
--- 0002out.csv
--- ...
</pre></pre>
各子文件与文件夹的说明如下：
1. **sdwpf_245days_v1.csv**：该数据集为2022年KDD Cup挑战赛发布的数据集，涵盖245天的观测数据。
2. **sdwpf_baidukddcup2022_turb_location.csv**：该文件提供了数据集中所有风电机组的相对位置信息。
3. **final_phase_test**：该数据集为2022年百度KDD Cup决赛阶段的测试数据，可用于对比参赛模型与2022年KDD Cup获奖队伍的算法效果。其包含`infile`文件夹与`outfile`文件夹：前者存放模型的输入数据，后者存放对应输出的真实标签（基准真值）。简言之，对于模型函数y = f(x)，x对应`infile`文件夹中的文件，y的真实值则对应`outfile`文件夹中的文件，例如`001out = f(001in)`。
关于2022年百度KDD Cup所用sdwpf_kddcup数据集的更多细节，请参考百度KDD Cup论文：《SDWPF：面向2022年KDD Cup空间动态风电功率预测挑战赛的数据集》。
**sdwpf_full**
sdwpf_full数据集相比2022年KDD Cup发布的版本包含更多信息，不仅涵盖SCADA（监控与数据采集）数据，还包含源自欧洲中期天气预报中心（European Centre for Medium-Range Weather Forecasts, ECMWF）第五代全球气候大气再分析数据集（ERA5）的气象数据，例如相对湿度、风速与风向。该数据集采集自一座包含134台风电机组的风电场，历时两年，数据覆盖2020年1月至2021年12月。其目录结构如下：
<pre><pre>sdwpf_full
--- sdwpf_turb_location_elevation.csv
--- sdwpf_2001_2112_full.csv
--- sdwpf_2001_2112_full.parquet
</pre></pre>
各文件说明如下：
1. **sdwpf_turb_location_elevation.csv**：该文件详细记录了数据集中所有风电机组的相对位置与海拔高度信息。
2. **sdwpf_2001_2112_full.csv**：该数据集涵盖2020年1月至2021年12月共两年的观测数据，采集自一座包含134台风电机组的风电场。相比sdwpf_kddcup/sdwpf_245days_v1.csv，该数据集做了全面优化：
- 时间跨度更长：覆盖2020年1月至2021年12月共两年，而sdwpf_245days_v1.csv仅涵盖245天的数据；
- 气象信息更丰富：新增了源自欧洲中期天气预报中心（ECMWF）第五代全球气候大气再分析数据集（ERA5）的相对湿度、风速、风向等气象数据；
- 时间细节更完整：相较于KDD Cup赛事阶段为防止数据关联而隐藏时间戳的设置，该版本为每条数据都添加了具体的时间戳信息。
3. **sdwpf_2001_2112_full.parquet**：该数据集与sdwpf_2001_2112_full.csv内容完全一致，仅数据存储格式不同。