AutoBEE: A hierarchical multi-agent approach for energy and environmental parameter analysis

Abstract

Traditional approaches to building performance analysis often rely on manual or semi-manual methods, suffering from cumbersome workflows, low efficiency, and high error rates. This study developed AutoBEE, an automated analysis framework for building energy consumption and environmental parameters based on hierarchical multi-agent system integrated with large language models. This framework focused on enhancing the efficiency of individual agent and productive collaboration among agents group. Through the development of a comprehensive agent tool library, the establishment of a multi-level network spanning from teams to agents, the design of a lightweight communication protocol, and the creation of dynamic path planning, AutoBEE achieves autonomous unmanned operation from natural language input to building performance report output. During operation, agents group can exhibit a wide range of capabilities, including but not limited to accurately parsing user instructions, decomposing complex tasks into subtasks, adjusting parameters in the input data file, executing simulation calculations by energy simulation software, and generating structured professional reports. Verified through experiments in 54 typical scenarios, compared with traditional methods, AutoBEE has significantly improved efficiency, accuracy, rationality, content richness, and economic feasibility, providing an innovative solution for building performance research.