Scholarlink

<p>Accurate electricity demand forecasting is essential for reliable power system operation, energy planning, and infrastructure development. In Saudi Arabia, electricity consumption is strongly affected by high temperatures, seasonal activities, religious events, and rapid industrial growth under Vision 2030. Traditional forecasting methods often struggle to capture the nonlinear and complex behaviour of regional electricity demand. This paper presents a comparative evaluation of four advanced ensemble and gradient boosting models, including CatBoost, XGBoost, LightGBM, and Random Forest, for electricity demand forecasting in the Makkah and Madinah regions of Saudi Arabia. A real long-term historical dataset spanning nearly two decades was used with climate-related and calendar-based features to model load variability. Model performance was evaluated using MAE, RMSE, MAPE, and <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation>. The results show that CatBoost achieved the best overall performance with an RMSE of 2029.85 GWh (out of a mean monthly consumption of approximately 37,800 GWh), MAE of 1458.10 GWh, MAPE of 3.86%, and an <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> of 0.985. The findings confirm the effectiveness of CatBoost for electricity demand forecasting in climate-sensitive and event-driven regions.</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

A catboost-based framework for accurate electricity load forecasting in Saudi Arabia

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月14日

查看全文
  • Abdulaziz Almalaq

摘要

Accurate electricity demand forecasting is essential for reliable power system operation, energy planning, and infrastructure development. In Saudi Arabia, electricity consumption is strongly affected by high temperatures, seasonal activities, religious events, and rapid industrial growth under Vision 2030. Traditional forecasting methods often struggle to capture the nonlinear and complex behaviour of regional electricity demand. This paper presents a comparative evaluation of four advanced ensemble and gradient boosting models, including CatBoost, XGBoost, LightGBM, and Random Forest, for electricity demand forecasting in the Makkah and Madinah regions of Saudi Arabia. A real long-term historical dataset spanning nearly two decades was used with climate-related and calendar-based features to model load variability. Model performance was evaluated using MAE, RMSE, MAPE, and \(R^2\) . The results show that CatBoost achieved the best overall performance with an RMSE of 2029.85 GWh (out of a mean monthly consumption of approximately 37,800 GWh), MAE of 1458.10 GWh, MAPE of 3.86%, and an \(R^2\) of 0.985. The findings confirm the effectiveness of CatBoost for electricity demand forecasting in climate-sensitive and event-driven regions.