Scholarlink

<p>To determine the most suitable data sources for wind power forecasting, an investigation was conducted employing machine learning algorithms. Specifically, XGBoost, random forest, and support vector machine algorithms were utilized for day-ahead predictions. These algorithms were selected due to the requirement in power exchange markets to estimate power generation one day prior to the power trading day. The predictive performance was assessed using actual power output data from a 2 MW wind turbine, gathered via a supervisory control and data acquisition system. The data sources included met mast, ground Light detection and ranging (LiDAR), wind turbine nacelle, and weather research and forecasting (WRF) model data. Training datasets spanning one month, three months, and six months were used to analyze the impact of the training period on prediction accuracy. Results indicated that day-ahead predictions based on met mast and LiDAR data outperformed those using nacelle and WRF data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of multi-source data for improvement of day-ahead wind power forecasting with machine learning models

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Byeongtaek Kim,
  • Kyungnam Ko

摘要

To determine the most suitable data sources for wind power forecasting, an investigation was conducted employing machine learning algorithms. Specifically, XGBoost, random forest, and support vector machine algorithms were utilized for day-ahead predictions. These algorithms were selected due to the requirement in power exchange markets to estimate power generation one day prior to the power trading day. The predictive performance was assessed using actual power output data from a 2 MW wind turbine, gathered via a supervisory control and data acquisition system. The data sources included met mast, ground Light detection and ranging (LiDAR), wind turbine nacelle, and weather research and forecasting (WRF) model data. Training datasets spanning one month, three months, and six months were used to analyze the impact of the training period on prediction accuracy. Results indicated that day-ahead predictions based on met mast and LiDAR data outperformed those using nacelle and WRF data.