Scholarlink

<p>This study aims to forecast high-frequency electricity prices using a hybrid machine learning decomposition model. Incorporating four types of weather data—temperature, humidity, wind gusts, and wind speed—the model predicts electricity prices at a fifteen-minute frequency in the Belgian market. Initially, a combination of the Reverse Unrestricted-MIDAS (RU-MIDAS) model and machine learning is tested but fails to produce satisfactory predictive outcomes. To address the complex characteristics of electricity price data and mitigate data leakage, a new forecasting framework, “Mixed-Frequency Rolling Singular Spectrum Analysis with Machine Learning” (MF-RSSA-ML), is proposed. An empirical analysis is conducted on three rolling decomposition algorithms. Rolling Empirical Ensemble Mode Decomposition performs poorly in both error metrics and the Diebold-Mariano (DM) test, while rolling Empirical Mode Decomposition improves error metrics but still fails the DM test. In contrast, the “MF-RSSA-ML” framework demonstrates superior predictive performance, improving MAE, RMSE, and SMAPE by up to 31.43%, 26.04%, and 20.79%, respectively. The results of the DM test provide additional evidence supporting the superiority of MF-RSSA-ML compared to the other two decomposition approaches. By effectively integrating weather-related influences, the “MF-RSSA-ML” system provides accurate high-frequency electricity price forecasts. The findings offer valuable insights for generators, wholesalers, and consumers in optimizing electricity production and consumption decisions.</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

Forecasting high-frequency electricity price with hybrid machine learning decomposition model

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Zhenpeng Tang,
  • Liuqing Wang,
  • Yi Cai,
  • Mohammad Zoynul Abedin

摘要

This study aims to forecast high-frequency electricity prices using a hybrid machine learning decomposition model. Incorporating four types of weather data—temperature, humidity, wind gusts, and wind speed—the model predicts electricity prices at a fifteen-minute frequency in the Belgian market. Initially, a combination of the Reverse Unrestricted-MIDAS (RU-MIDAS) model and machine learning is tested but fails to produce satisfactory predictive outcomes. To address the complex characteristics of electricity price data and mitigate data leakage, a new forecasting framework, “Mixed-Frequency Rolling Singular Spectrum Analysis with Machine Learning” (MF-RSSA-ML), is proposed. An empirical analysis is conducted on three rolling decomposition algorithms. Rolling Empirical Ensemble Mode Decomposition performs poorly in both error metrics and the Diebold-Mariano (DM) test, while rolling Empirical Mode Decomposition improves error metrics but still fails the DM test. In contrast, the “MF-RSSA-ML” framework demonstrates superior predictive performance, improving MAE, RMSE, and SMAPE by up to 31.43%, 26.04%, and 20.79%, respectively. The results of the DM test provide additional evidence supporting the superiority of MF-RSSA-ML compared to the other two decomposition approaches. By effectively integrating weather-related influences, the “MF-RSSA-ML” system provides accurate high-frequency electricity price forecasts. The findings offer valuable insights for generators, wholesalers, and consumers in optimizing electricity production and consumption decisions.