Towards Efficient Building Designing: Heating and Cooling Load Prediction via Multi-Output Model.

Area of Science:

• Building science and sustainable architecture
• Machine learning applications in energy systems
• Environmental engineering and computational modeling

Background:

• Growing concerns over energy consumption and environmental impact necessitate energy-efficient building design.
• Building layout factors like compactness, area, height, orientation, and glazing significantly influence cooling load (CL) and heating load (HL).
• Accurate CL and HL prediction is crucial for effective energy management and improved inhabitant comfort.

Purpose of the Study:

• To address limitations of traditional single-output machine learning (ML) models for CL and HL prediction.
• To develop a robust and generalized framework for concurrent prediction of CL and HL.
• To introduce a novel multi-output (MO) sequential learning model for energy efficiency applications.

Main Methods:

• Proposed a novel framework utilizing a gated recurrent unit (GRU) for sequential learning.
• Implemented a multi-output (MO) strategy to predict CL and HL concurrently within a unified model.
• Incorporated utility preprocessing and conducted comprehensive ablation studies comparing ML and deep learning (DL) techniques.

Main Results:

• The proposed GRU-based MO framework demonstrated superior performance in predicting CL and HL compared to existing models.
• The unified framework effectively handled the nonlinearity between input factors and output loads.
• Ablation studies confirmed the robustness and generalization capabilities of the novel approach.

Conclusions:

• The developed GRU-based MO model offers a significant advancement for accurate and concurrent prediction of building energy loads.
• This approach enhances energy management strategies and living standards in residential buildings.
• The study pioneers the application of MO sequential learning with utility preprocessing in a unified framework for energy efficiency.