Multi-objective optimization of shading devices using ensemble machine learning and orthogonal design of experiments

Abstract

Buildings are significant sources of energy consumption due to the non-optimal configuration of shading devices that results in excessive solar heat gain. Amorphous optimization of shading devices has not been explored in-depth using sophisticated modelling tools despite research efforts to optimize shading devices for different objectives such as energy consumption and occupant comfort. This study introduces updates to the existing framework for optimizing shading devices for multiple objectives. The objectives include occupants’ thermal Discomfort Hours (DH), Heating and Cooling loads (H/C), and total shading devices’ surface area. The existing framework in literature follows three main phases including data generation, Machine Learning (ML) models development, and optimization. In this study, three updates to the mentioned framework are proposed: First, the optimization of fixed shading devices is performed amorphously for maximizing the occupants’ thermal comfort; Second, the use of orthogonally structured training dataset to develop ML models; Third, incorporating the modern technique of Multi-Objective Manta-Ray Foraging Optimizer (MOMRFO) in the optimization phase. For only perimeter zones, amorphous shading devices can incur a maximum 12.7 % of H/C compared to the present situation of the case study when optimized for the H/C objective, while a maximum 2.8 % of DH can be saved compared to the present situation of the case study when optimized for the DH objective. Also, it is found that the payback period becomes high (18.35 years) when the DH objective is prioritized over other objectives. The study introduces a framework that couples orthogonal design of experiment with ensemble ML models to optimize the geometrical configuration of amorphous shading devices.