A multi-objective optimization strategy for ultra-low energy residential buildings based on a hybrid machine learning algorithm

To address the lack of effective multi-objective optimization approaches that balance building energy consumption, indoor thermal comfort, and life-cycle cost in ultra-low energy residential buildings, this study proposes a hybrid algorithm that integrates the Non-dominated Sorting Genetic Algorithm-II (NSGA-II) with Multi-Objective Particle Swarm Optimization (MOPSO) for the multi-objective optimization design of such buildings, validated through case studies in two climate zones. Orthogonal experiments were conducted on eight variables, including the heat transfer coefficients of external walls, roofs, and windows; directional window-to-wall ratios; and solar heat gain coefficients. An artificial neural network model for the three objectives was developed to serve as fitness functions for the hybrid algorithm. The hybrid approach outperformed the standalone NSGA-II and MOPSO in terms of the hypervolume, inverted generational distance, and spacing metrics, demonstrating superior convergence and solution diversity. Optimized design parameter ranges were derived, and best solutions were identified for both climates, providing practical guidance for similar regions. The innovations of this study include: (1) a multi-objective optimization framework balancing energy, comfort, and cost to enhance solution practicality in ultra-low energy residential buildings; (2) integration of NSGA-II-MOPSO with metamodeling for the three objectives, verified to improve optimization efficiency over individual algorithms.