Short-term wind power forecasting methods based on machine learning: A review and case study
Xiaojie Guo, Pingliang Zeng, Xiong Xiong, Guangwei Wang, Yang Cui
Abstract
Accurate wind power prediction is crucial for the efficient and optimized operation of power systems with high renewable energy penetration. As global installed wind power capacity continues to grow, vast amounts of meteorological and wind turbine operation data are collected. Consequently, data-driven machine learning methods have become widely adopted for power prediction. This paper provides a comprehensive review of short-term wind power prediction methods based on machine learning, detailing the modeling process from a machine learning perspective. The paper outlines the key steps in the process, including data preprocessing, feature engineering, model selection, hyperparameter optimization, and evaluation. A unified modeling framework is proposed, where power prediction models are constructed by selecting one or more base methods from each step, which are then integrated using a machine learning-based approach. Additionally, from the perspective of feature engineering, the paper introduces four model integration paradigms, addressing feature mining, feature selection, feature decomposition, and feature fusion, to guide the design of integrated models. To validate the proposed framework, 2025 integrated power prediction models are designed and tested using data from wind farms in mountainous regions of China, demonstrating the feasibility and effectiveness of the framework. Finally, the paper outlines four promising directions for future research in wind power forecasting.