Preprocessed Spatial Dynamic Wind Power Forecasting (SDWPF) Dataset from Baidu KDD Cup 2022 Challenge

This repository contains a preprocessed dataset derived from the Spatial Dynamic Wind Power Forecasting (SDWPF) challenge, hosted by Baidu KDD Cup 2022. The original challenge aimed to facilitate the progress of data-driven machine learning methods for wind power forecasting.The raw data underwent a comprehensive preprocessing pipeline to enhance its quality and applicability for machine learning tasks. Key preprocessing steps include:1. **Basic Preprocessing:** Handling invalid data (e.g., negative power, zero power with high wind speed, abnormal pitch/nacelle angles), filling missing values (bfill then ffill), and engineering time-related features (Day, Hour, Minute) from 'Tmstamp'. A new feature 'Pab_max' (maximum of Pab1, Pab2, Pab3) was created, and original pitch angle columns were dropped.2. **Wavelet Denoising:** Application of Discrete Wavelet Transform (db4 wavelet, level 1, soft thresholding) to reduce noise and highlight underlying trends in the time-series data.3. **Standardization:** Independent standardization of features and the target variable ('Patv') using StandardScaler, with the scalers saved for consistent future use.The original dataset fields included: TurbID, Day, Tmstamp, Wspd, Wdir, Etmp, Itmp, Ndir, Pab1, Pab2, Pab3, Prtv, Patv.The final features used for model input after preprocessing are: Pab_max, Wspd, Wdir, Etmp, Itmp, Ndir, Prtv, Patv, Day, Hour, Minute.This processed dataset is suitable for research and development in wind power forecasting models, offering a clean and standardized input for machine learning algorithms.