Wind Power Prediction Based on Data Cleaning and Feature Analysis

Accurate forecasting of wind energy is essential for ensuring the reliable functioning of wind power systems. However, data collected during acquisition and transmission is prone to issues, and abnormal events can degrade the quality of wind power data, thereby affecting prediction accuracy. This paper delves into data cleaning and feature analysis to improve the precision of wind power forecasts. In the data cleaning phase, this paper introduces a method for cleaning wind farm data utilizing QM-DBSCAN, which effectively handles various problematic data during data acquisition and transmission, as well as data affected by abnormal events, significantly improving the accuracy of data cleaning. In terms of feature analysis, SHAP analysis is employed to extract features from the cleaned data. By quantifying the impact of each feature on the prediction results through SHAP values, key features are identified, not only validating the effectiveness of data cleaning but also optimizing the structure of the prediction model by removing redundant features. In the end, A transformer model is employed to forecast wind power. This study uses real operational data from a wind farm located in South China as a case study, creates a scatter plot of wind speed versus power, and applies an anomaly detection technique that combines the quartile method with DBSCAN clustering to enhance the precision of anomaly identification. Tests indicate that the suggested approach significantly improves the precision of wind power forecasts, offering dependable technical backing for wind power prediction.