A short-term load forecasting framework for air conditioning system based on model stacking.

Area of Science:

• Energy Management
• Artificial Intelligence
• Building Systems

Background:

• Accurate short-term air conditioning load forecasting is essential for optimizing energy consumption and enabling intelligent control of HVAC systems.
• Existing forecasting methods may not fully capture the complex dynamics of air conditioning loads, necessitating advanced approaches.

Purpose of the Study:

• To propose and validate a novel short-term air conditioning load forecasting framework utilizing model stacking.
• To enhance prediction accuracy and efficiency for air conditioning energy management through ensemble learning techniques.

Main Methods:

• Developed a forecasting framework based on model stacking, integrating six machine learning models: Lasso regression, Ridge regression, Random Forest, Support Vector Regression, eXtreme Gradient Boosting, and Long Short-Term Memory.
• Implemented operational procedures including feature screening, hyperparameter optimization, and cross-stacking of prediction models.
• Validated the framework on a real-world air conditioning system.

Main Results:

• The proposed model stacking framework achieved improved prediction accuracy, with 28 out of 36 control simulations showing enhanced results.
• An average increase in R-squared of 6.4% was observed, demonstrating the effectiveness of the ensemble approach.
• Simpler submodels within the meta-model generally yielded better performance, while complex coupling models could degrade accuracy.

Conclusions:

• The model stacking framework offers a practical and effective solution for short-term air conditioning load forecasting, balancing accuracy with computational efficiency.
• Findings provide valuable guidance on implementing model stacking and selecting appropriate meta-models for HVAC energy prediction.
• The approach contributes to the advancement of intelligent building energy management systems.