Extracting Heuristics for Designing Sustainable Built Environments by Coupling Multiobjective Evolutionary Optimization and Machine Learning

Abstract

Abstract In sustainable process design, computational optimization methods are often used to find a single or a set of Pareto optimal or near-optimal solutions. However, knowledge about what makes the solutions optimal can be more valuable as it can be formulated as heuristics for guiding future designs of similar systems. In this paper, we propose a methodology that couples multiobjective optimization and machine learning to extract design heuristics. The methodology has been demonstrated by a case study on sustainable residential system design. Potential heuristics for sustainable buildings include: 1) keep the house as warm as comfortable in the summer and as cold as comfortable in the winter; 2) place vegetable garden, shading trees, and rain water harvesting system around the house. This further indicates the benefits of the synergistic design between technologies and ecosystems. The approach of combining optimization with learning is general enough to be applied to other systems, ranging from local manufacturing processes to urban systems.