Scholarlink

With the building sector accounting for 40% of global energy consumption, achieving Net Zero by 2050 emerges as a paramount challenge, necessitating precise energy consumption forecasting to streamline smart energy supply chains amidst unforeseen uncertainties. Traditional prediction models often falter in navigating the intricate, non-linear interplay of factors such as climate, thermal system performance, and occupancy behaviours, creating a critical research gap. This study introduces a groundbreaking hybrid machine-learning algorithm that synergizes the sparse, interpretable, and transparent (SIT) nature of the NARMAX model’s advanced temporal sequence processing capabilities. Employing the REFIT Smart Home dataset, which provides two years of hourly resolution data, our methodology showcases remarkable precision in probabilistic energy consumption forecasting. A comparative analysis underscores the proposed hybrid model’s superiority over established methods, including Artificial Neural Networks (ANN), Support Vector Machines (SVM), particularly in reducing Root Mean Squared Error (RMSE) and improving the Coefficient of Variation (CV). This innovative fusion not only bridges the existing precision-interpretability gap but also paves the way for more efficient, predictive energy management frameworks in the building sector.

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

Development of SIT Hybrid Machine Learning Algorithm for Hour Level Building Energy Consumption Probabilistic Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yiping Meng,
  • Yiming Sun,
  • Sergio Rodriguez,
  • Farzad Pour Rahimian

摘要

With the building sector accounting for 40% of global energy consumption, achieving Net Zero by 2050 emerges as a paramount challenge, necessitating precise energy consumption forecasting to streamline smart energy supply chains amidst unforeseen uncertainties. Traditional prediction models often falter in navigating the intricate, non-linear interplay of factors such as climate, thermal system performance, and occupancy behaviours, creating a critical research gap. This study introduces a groundbreaking hybrid machine-learning algorithm that synergizes the sparse, interpretable, and transparent (SIT) nature of the NARMAX model’s advanced temporal sequence processing capabilities. Employing the REFIT Smart Home dataset, which provides two years of hourly resolution data, our methodology showcases remarkable precision in probabilistic energy consumption forecasting. A comparative analysis underscores the proposed hybrid model’s superiority over established methods, including Artificial Neural Networks (ANN), Support Vector Machines (SVM), particularly in reducing Root Mean Squared Error (RMSE) and improving the Coefficient of Variation (CV). This innovative fusion not only bridges the existing precision-interpretability gap but also paves the way for more efficient, predictive energy management frameworks in the building sector.