Scholarlink

Buildings account for nearly 40% of global energy consumption, with cooling demands dominating electricity consumption in semi-arid climates. Therefore, accurate short-term forecasting of building energy consumption is essential for improving energy efficiency and supporting sustainable urban development. This study investigates the application of artificial intelligence (AI) methods for electricity demand forecasting in semi-arid office buildings, using the Building Data Genome Project 2 (BDG2) dataset and three temporal input scenarios that incorporate calendar data. Four models were tested and evaluated: a baseline Artificial Neural Network (ANN), a Long Short-Term Memory (LSTM) network, an XGBoost model, and a hybrid Convolutional Neural Network with LSTM (CNN–LSTM). Models performance was evaluated using RMSE, MAE, and MAPE, and the results show that XGBoost achieved the best evaluation metrics amongst the other models, while ANN had the weakest ones. The inclusion of weekday/weekend differentiation improved accuracy in some models, while additional calendar features provided limited additional benefit. These findings demonstrate the effectiveness of tree-based models for energy consumption forecasting in smart buildings and support their integration into Building Management Systems (BMS) to advance sustainable energy optimization strategies, particularly in a semi-arid context.

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

Weather-Aware Machine Learning Models for Energy Consumption Forecasting of Buildings in Semi-Arid Climate Zones

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Iguerd Nouhaila,
  • Elassali Raja,
  • El kabtane Hamada,
  • Nefzaoui Elyes

摘要

Buildings account for nearly 40% of global energy consumption, with cooling demands dominating electricity consumption in semi-arid climates. Therefore, accurate short-term forecasting of building energy consumption is essential for improving energy efficiency and supporting sustainable urban development. This study investigates the application of artificial intelligence (AI) methods for electricity demand forecasting in semi-arid office buildings, using the Building Data Genome Project 2 (BDG2) dataset and three temporal input scenarios that incorporate calendar data. Four models were tested and evaluated: a baseline Artificial Neural Network (ANN), a Long Short-Term Memory (LSTM) network, an XGBoost model, and a hybrid Convolutional Neural Network with LSTM (CNN–LSTM). Models performance was evaluated using RMSE, MAE, and MAPE, and the results show that XGBoost achieved the best evaluation metrics amongst the other models, while ANN had the weakest ones. The inclusion of weekday/weekend differentiation improved accuracy in some models, while additional calendar features provided limited additional benefit. These findings demonstrate the effectiveness of tree-based models for energy consumption forecasting in smart buildings and support their integration into Building Management Systems (BMS) to advance sustainable energy optimization strategies, particularly in a semi-arid context.