Wind power data cleaning using RANSAC-based polynomial and linear regression with adaptive threshold.

Area of Science:

• Renewable Energy Systems
• Data Science and Analytics
• Power Systems Engineering

Background:

• Rising global demand for clean energy highlights wind power's importance.
• Dense anomalies in wind power data degrade forecasting accuracy and jeopardize grid security.
• Existing data cleaning methods struggle with high proportions of dense anomalies.

Purpose of the Study:

• To propose an adaptive threshold robust regression model (RPR model) for effective wind power data cleaning.
• To address the challenge of dense anomalies in wind power datasets.
• To enhance the accuracy of wind power forecasting models and ensure grid security.

Main Methods:

• Developed an RPR model combining the Random Sample Consensus (RANSAC) algorithm and polynomial linear regression.
• Extended polynomial features to capture nonlinear relationships between wind speed and power.
• Employed dynamic threshold adjustment based on median of residuals and median absolute deviation (MAD) for anomaly detection and cleaning.

Main Results:

• The RPR model demonstrated superior performance in data cleaning compared to existing methods (Bidirectional Change Point Grouping Quartile Statistical Model, Principal Contour Image Processing Model, DBSCAN, SVM).
• Achieved a significant 72.1% reduction in the average absolute error (MAE) of wind power forecasting models.
• Effectively reduced prediction errors for Convolutional Neural Network (CNN) + Gated Recurrent Unit (GRU) forecasting models, ensuring high accuracy.

Conclusions:

• The proposed adaptive threshold robust regression model is an innovative and effective approach for wind power data cleaning.
• The model excels in handling datasets with high proportions of dense anomalies, outperforming conventional methods.
• This method offers significant potential for improving wind power data quality, forecasting accuracy, and grid security.