Scholarlink

<p>Electricity consumption is recognized as one of the fundamental indicators of economic activities. Accurate forecasting of electricity prices is therefore critical for economic planning and sustainable development. This research investigates the effectiveness of 19 different machine learning algorithms/models in forecasting the US electricity prices. It provides a comprehensive analysis by evaluating approaches that both include and exclude seasonality factors. Key findings reveal that (i) CatBoost Regressor with TimeSeries strategy achieved the highest forecasting accuracy with a mean absolute error (MAE) of 0.0081 and root mean squared logarithmic error (RMSLE) of 0.0093; (ii) Gradient Boosting Regressor and CatBoost Regressor through k-fold strategy revealed almost the best similar forecast accuracies, (iii) when incorporating seasonality, with MAE = 0.0079, and RMSLE = 0.0091, Linear Regression and other linear models (Ridge, Bayesian Ridge) outperformed tree-based models highlighting the importance of explicit seasonal features for linear methods, (iv) models trained with TimeSeries outperformed k-fold due to better alignment with temporal structure; and (v) accounting for seasonal trends improved prediction accuracy across most models. This study highlights the effectiveness of ML models, especially CatBoost and Linear Regression with seasonality in forecasting electricity prices. The findings offer valuable guidance for policymakers, energy analysts, and investors seeking to enhance strategic planning in the energy sector.</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 electricity price index with machine learning models and strategies

  • 期刊论文

  • 发表日期：

    2025年9月13日

查看全文
  • Sevda Kuşkaya,
  • Faik Bilgili

摘要

Electricity consumption is recognized as one of the fundamental indicators of economic activities. Accurate forecasting of electricity prices is therefore critical for economic planning and sustainable development. This research investigates the effectiveness of 19 different machine learning algorithms/models in forecasting the US electricity prices. It provides a comprehensive analysis by evaluating approaches that both include and exclude seasonality factors. Key findings reveal that (i) CatBoost Regressor with TimeSeries strategy achieved the highest forecasting accuracy with a mean absolute error (MAE) of 0.0081 and root mean squared logarithmic error (RMSLE) of 0.0093; (ii) Gradient Boosting Regressor and CatBoost Regressor through k-fold strategy revealed almost the best similar forecast accuracies, (iii) when incorporating seasonality, with MAE = 0.0079, and RMSLE = 0.0091, Linear Regression and other linear models (Ridge, Bayesian Ridge) outperformed tree-based models highlighting the importance of explicit seasonal features for linear methods, (iv) models trained with TimeSeries outperformed k-fold due to better alignment with temporal structure; and (v) accounting for seasonal trends improved prediction accuracy across most models. This study highlights the effectiveness of ML models, especially CatBoost and Linear Regression with seasonality in forecasting electricity prices. The findings offer valuable guidance for policymakers, energy analysts, and investors seeking to enhance strategic planning in the energy sector.