Paper Conference

Proceedings of eSim 2020: 11th Conference of IBPSA-Canada


Multi-Stage Agent-Based Optimization of Building Layouts for Energy Performance

David Rulff, Ralph Evins
University of Victoria, Canada

Abstract: Early decisions on building program and room topology influence built form and zoning, often imposing immutable constraints on system design while having a disproportionately large impact on performance. This paper explores analogous concepts and techniques from a variety of fields (including graph theory, architectural practice, machine learning and physics simulation) to develop a novel agent-based multi-objective method to rapidly generate feasible building layouts and converge on potentially optimal solutions. The problem is divided into three stages, each incorporating performance objectives and zoning constraints: room positioning (pseudo-physics-based dense packing of undirected, weighted graphs), layout generation (cellular growth), and performance refinement. An illustrative example for the first two stages is provided, with refinement options discussed at a qualitative level. Applying this method during the early conceptualization of buildings would support more integrated design processes targeting ambitious performance targets. The intention is to provide insight into the problem and directions for future work.