Multi-Objective Optimization and Machine Learning Approaches for Low-Carbon Building Design

Abstract

The building sector accounts for approximately 40% of global energy-related carbon dioxide emissions, making it a critical focus area for climate change mitigation. As the world advances toward carbon neutrality goals, low-carbon building design has emerged as an essential strategy combining environmental sustainability with economic viability. This review examines recent developments in multi-objective optimization approaches and machine learning techniques for low-carbon building design, covering the period from 2018 to 2025. We synthesize research on optimization algorithms including NSGA-II, SPEA2, and hybrid evolutionary approaches, alongside machine learning methods such as neural networks, random forests, and support vector machines. The review identifies three primary optimization objectives: carbon emissions reduction, cost minimization, and building performance enhancement including energy efficiency, daylighting, and thermal comfort. Key findings reveal that integrating Building Information Modeling with parametric design and surrogate modeling can reduce computational time by up to 50% while achieving 13 to 25% reductions in carbon emissions. However, significant challenges remain in data quality, model interpretability, and the integration of embodied and operational carbon across building lifecycles. This review provides a comprehensive framework for researchers and practitioners, identifying critical gaps and future research directions toward achieving net-zero buildings.