Scholarlink

This study builds and compares the effectiveness of a series of forecasting models to find the optimal solution for the short-term electricity load fore-casting problem in 21 provinces in southern Vietnam (excluding Ho Chi Minh City). The context of the increasing demand for electricity consumption, influenced by complex factors such as weather, socio-economic conditions, and unusual events, requires forecasting models to have high accuracy and adaptability. The study proposes implementing the Rolling_SARIMAX model using the rolling forecast method, where the model is continuously updated with the latest data to enhance its adaptability to real-world fluctuations. The temperature variable and calendar features are integrated as exogenous variables to enhance accuracy. The performance of this model is comprehensively compared with traditional statistical models (ARIMA, SARIMA, Prophet), modern deep learning models (LSTM, GRU, attLSTM), and two-stage hybrid models (SARIMAX + attLSTM, SARIMAX + XGBoost). Experimental results on daily electricity load data from 2021–2024 show that the Rolling_SARIMAX model achieved outstanding performance with a mean absolute percentage error (MAPE) of only 1.78%. The study confirms that a linearly deployed model with a dynamic updating mechanism can achieve higher efficiency than complex deep learning architectures and proposes the application of Rolling_SARIMAX in power dispatch to minimize risks and enhance system operational efficiency.

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

Enhanced Short-Term Electricity Load Forecasting in Southern Vietnam Using a Rolling SARIMAX Framework with Exogenous Variable Integration

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Long Kieu Hoang,
  • Vinh Nguyen Duc,
  • Hien Vo Thi Ngoc,
  • Kien Le Trung,
  • Hoan Bui Quoc,
  • Huong Bui

摘要

This study builds and compares the effectiveness of a series of forecasting models to find the optimal solution for the short-term electricity load fore-casting problem in 21 provinces in southern Vietnam (excluding Ho Chi Minh City). The context of the increasing demand for electricity consumption, influenced by complex factors such as weather, socio-economic conditions, and unusual events, requires forecasting models to have high accuracy and adaptability. The study proposes implementing the Rolling_SARIMAX model using the rolling forecast method, where the model is continuously updated with the latest data to enhance its adaptability to real-world fluctuations. The temperature variable and calendar features are integrated as exogenous variables to enhance accuracy. The performance of this model is comprehensively compared with traditional statistical models (ARIMA, SARIMA, Prophet), modern deep learning models (LSTM, GRU, attLSTM), and two-stage hybrid models (SARIMAX + attLSTM, SARIMAX + XGBoost). Experimental results on daily electricity load data from 2021–2024 show that the Rolling_SARIMAX model achieved outstanding performance with a mean absolute percentage error (MAPE) of only 1.78%. The study confirms that a linearly deployed model with a dynamic updating mechanism can achieve higher efficiency than complex deep learning architectures and proposes the application of Rolling_SARIMAX in power dispatch to minimize risks and enhance system operational efficiency.