Intelligent Agent for Construction Safety Reporting BasedonLarge Language Models

In the construction industry, construction safety management is primarily governed by asafetymanagement team that involves safety of icers, supervisors, project manager, and owner'srepresentatives. Safety reporting plays a vital role in safety management, serving not only to document on-site activities but also to facilitate ef ective communication across various management levels. With digital-technology advancements, intelligent safety reporting systems based on electronic formshave emerged. However, these systems encounter challenges such as high learning costs, delays inreport submission and unstable report quality. This study presents an intelligent agent designedforsafety reporting utilizing large language model (LLM) technology. The agent is deployed oncloudserver, and safety personnel can connect with it through an instant messaging application toobtainreal-time information. The interaction between the agent and safety personnel is similar todailyconversations, requiring minimal learning costs. Additionally, the agent seamlessly generates safetyreports, of ering a concise summary and subsequent work plans, thereby improving the overall qualityof reports. Through its deployment on a construction site, the agent achieved an F1 score of 0.996forhazard identification and an average score of 93.4 for generated reports. It has been recognizedbyconstruction safety personnel, proving its ability to improve the ef iciency and quality of safetypersonnel's work and demonstrating its potential to enhance safety management capabilities.