Customized large-scale model for human-AI collaborative operation and maintenance management of building energy systems

Artificial intelligence (AI) is becoming an integral part for operation and maintenance (O&M) management, propelling the low-carbon transition of building energy systems. However, as the users of building energy systems, human beings are detached from the decision-making loop of AI, which leads to the suboptimal performance for O&M tasks in practical applications. To this end, we proposed a customized large-scale model for human-AI collaborative O&M management in building energy systems. The human-AI collaboration mechanism is characterized by humans providing the domain knowledge and specialized tools to guide AI while AI performing tasks to serve human needs. Specifically, the few-shot learning has been utilized in prompt engineering for the customized large-scale model to route various O&M tasks to corresponding solution paths: direct response, retrieving knowledge or invoking tools, which increases the accuracy of task routing from 73.7 % to 90.3 %. For specialized O&M tasks requiring domain knowledge, the multiple semantic levels of energy-specific knowledge are extracted by recursive clustering and integrated into the customized large-scale model by retrieval-augmented generation. The expert questionnaire indicates that the customized large-scale model outperforms ChatGPT-4o in 84 % of survey questions and generates more accurate and concise responses than GLM-4-9B. For complex O&M tasks requiring modeling and computation, the customized large-scale model is integrated with the specialized tools through data-structuring transformation aided by self-supervised reconstruction. The experimental test indicates that the accuracy of the customized large-scale model for fault diagnosis is 96.3 %, which outperforms general large-scale models by over 50 %. Our study will contribute to the human-AI collaboration for more efficient and safety O&M management, thereby accelerating the pace towards net zero emissions in building energy systems.