2026-03-23 パシフィック・ノースウェスト国立研究所(PNNL)
BEM-AI, a new agentic AI tool from PNNL, can help speed up the process of modeling energy consumption in commercial buildings.(Illustration by Cortland Johnson | Pacific Northwest National Laboratory)
＜関連情報＞
- https://www.pnnl.gov/news-media/new-ai-bot-offers-speedy-revenue-saving-building-energy-modeling
- https://www.sciencedirect.com/science/article/abs/pii/S0378778825014422
建物のエネルギーモデリングのための動的マルチエージェントネットワークの開発：拡張性と自律性を備えたエネルギーモデリングに向けたケーススタディ Development of a dynamic multi-agent network for building energy modeling: A case study towards scalable and autonomous energy modeling
Weili Xu, Hanlong Wan, Supriya Goel, Chrissi A. Antonopoulos
Energy and Buildings Available online: 19 November 2025
DOI:https://doi.org/10.1016/j.enbuild.2025.116712
Highlights
- Introduces BEM-AI: a scalable, modular framework for Building Energy Modeling (BEM) workflow.
- Leverages A2A & MCP protocols for interoperable AI collaboration in BEM tools.
- Robust automation of modeling tasks validated through case studies and reliability tests.
- Highlights BEM-AI’s adaptability and extensibility for future AI-driven BEM workflows.
Abstract
Building Energy Modeling (BEM) has progressed from simple thermal load calculations to a sophisticated tool for optimizing energy performance, improving occupant comfort, and supporting renewable energy integration. Despite this evolution, its broader adoption in practice remains constrained by the significant expertise, cost, and time required to develop reliable models. Recent advances in artificial intelligence (AI) present new opportunities to automate and streamline BEM workflows. This paper introduces AI-enabled building energy modeling (BEM-AI), a dynamic multi-agent framework that leverages structured agent coordination to address the limitations of traditional BEM approaches. Built on two foundational communication protocols, Agent-to-Agent (A2A) and Model Context Protocol (MCP), BEM-AI enables scalable, modular, and interoperable collaboration between specialized agents, tools, and external data resources. Unlike static multi-agent systems or monolithic fine-tuned models, BEM-AI dynamically decomposes modeling tasks into smaller, domain-specific components, allowing workflows to emerge adaptively rather than being pre-defined. Through the test cases, BEM-AI demonstrated the capacity to autonomously decompose user queries, yielding nearly 50% reductions in token consumption and execution time through context engineering with small language models. The test scenarios also reproduced typical energy performance outcomes such as a 0.8% reduction in energy use intensity (42.25 vs. 41.85 kBtu/ft2) from a 10% window-to-wall ratio decrease and a 2.3% reduction in whole-building electricity use (including an 8.5% reduction in interior lighting) from adding daylighting controls, illustrating the framework’s ability to leverage BEM software to automatically generate and evaluate standard modeling procedures. These results highlight BEM-AI’s efficiency in model orchestration and its potential to streamline the exploration of energy-saving strategies without relying on custom scripts or manual configuration.
