A research team led by Prof. HUANG Xinyan, Associate Professor of Department of Building Environment and Energy Engineering at The Hong Kong Polytechnic University (PolyU), has published a pioneering “LLM-AI” autonomous firefighting emergency response agent framework in Engineering, a flagship journal of the Chinese Academy of Engineering. The study presents a new solution for the future development of smart firefighting.
Powered by a Large Language Model (LLM) as its reasoning and task coordination core, the system integrates the ConvLSTM-Fire prediction model based on artificial intelligence of things (AIoT) technology. To overcome the constraints of manual data handling during emergencies, the system incorporates a self-driven mechanism. It can autonomously translate users’ natural-language requests related to fireground conditions and emergency response needs into executable tasks, generate and execute code, and automatically debug and retry upon encountering errors. This establishes a fully closed-loop workflow spanning data acquisition, model invocation, analytical reasoning, and result visualisation.
Experimental evaluations demonstrate that even under extreme fireground conditions where sensor failures occur, the system maintains a prediction accuracy of over 97% for two-dimensional temperature fields and smoke propagation pathways, exhibiting exceptional robustness. This research advances LLMs beyond conventional “static Q&A” interfaces, transforming them into intelligent firefighting emergency decision-making entities endowed with perception, computation and execution capabilities, thereby enhancing the usability and reliability of smart firefighting technologies in real-world fireground environments. Moving forward, the team aims to incorporate multimodal sensory data, strengthen cross-scenario generalisation capabilities, and accelerate the real-world deployment of this technology in smart buildings and municipal firefighting.
Prof. Huang is currently a member of the Otto Poon Charitable Foundation Smart Cities Research Institute (SCRI), the Otto Poon Charitable Foundation Research Institute for Smart Energy (RISE), the Research Institute for Sustainable Urban Development (RISUD), and the Research Centre for Resources Engineering towards Carbon Neutrality (RCRE) at PolyU.
Read the full paper: https://www.sciencedirect.com/science/article/pii/S2095809926001244?via%3Dihub
Online coverage:
Sciencenet - https://polyu.me/4yhr5Cj
黃鑫炎教授提出新型自驅動消防應變智能體 融合智慧火災預測與大語言模型
香港理工大學建築環境及能源工程系黃鑫炎副教授與團隊，於中國工程院院刊《Engineering》發表「LLM-AI」自驅動消防應急智能體框架，為未來智慧消防提出全新解決方案。
該系統以大語言模型作為推理與任務協調核心，整合基於人工智能物聯網技術的 ConvLSTM-Fire 火災預測模型。針對緊急情況下人工處理複雜數據的局限，系統引入自驅動機制，能依據使用者以自然語言提出的火場狀況查詢或應急決策需求，自主拆解任務、生成並運行程式碼，並且能在出錯時自動修正及重試，實現了從數據獲取、模型調用、分析推理到結果視覺化的全流程閉環。
實驗顯示，即使在火場極端條件環境下部分感測器失效，系統對二維溫度場與煙氣路徑的預測準確率仍高達97%以上，展現極高穩健性。此項研究推動大語言模型從傳統的「靜態問答」模式，升級為具備感知、計算與執行能力的「消防應急智能主體」，提升智慧消防技術在真實火場環境中的可用性與可靠性。未來，團隊將進一步整合多模態感知資訊，增強跨場景泛化能力，加速推動相關技術在智慧建築與城市消防領域的實際部署。
黃教授現為理大潘樂陶慈善基金智慧城市研究院、潘樂陶慈善基金智慧能源研究院、可持續城市發展研究院及碳中和資源工程研究中心成員。
閱讀研究全文 : https://www.sciencedirect.com/science/article/pii/S2095809926001244?via%3Dihub
網上報導：
Sciencenet - https://polyu.me/4yhr5Cj
