基于数据驱动的施工安全隐患特征网络分析与预警策略

Data-driven network analysis of construction hazard characteristics and warning strategy

建筑业施工安全隐患排查与治理不断向数字化和智能化转型,但目前对大规模且高维的安全隐患数据的高效解析仍然不够充分。该研究通过对施工现场安全隐患记录数据进行隐患特征提取和降维,构建隐患特征网络,并提出基于数据驱动的隐患预警策略。首先,通过文本挖掘技术对施工安全隐患记录进行标准化,提取出111个安全隐患特征,并对隐患特征进行层次聚类,形成11个隐患特征群;其次,通过相关性检验确定隐患特征之间的关联,进而计算出隐患特征关联强度,构建了安全隐患特征网络;进而,基于网络结构指标和个体指标分析,结合特征群聚类分析,辨识了关键安全隐患特征;最后,提出一种基于特征数据驱动的安全隐患预警策略,为更加高效地进行安全隐患排查治理、提升安全生产水平提供了参考。

[Objective] The investigation and management of construction hazards have transformed with the digitization and intellectualization of the construction industry. However, the efficient analysis and application of large-scale and high-dimensional hazard data remain farfetched. In this study, a hazard characteristic network is constructed by extracting and reducing the dimensions of the hazard record data on the construction site and data-driven hazard warning strategies are preseneted. [Methods] First, the hazard records are standardized using the text mining method and 111 safety hazard characteristics are extracted. Second, the correlations of these characteristics are determined using the correlation test, and the characteristic network is established based on the correlation strength calculation. Third, critical hazard characteristics are identified based on the analysis of the structural and individual indices in the network. In addition, 11 groups are obtained by the hierarchical clustering of all hazard characteristics. Finally, based on the above methods and analyses, data-driven warning strategies are preseneted in view of hazard characteristics. [Results] Through charateristic extraction and hierarchical clustering of potential safety hazards, 111 potential safety hazard characteristics and 11 potential safety hazard characteristic groups are determined to reduce the dimensions of potential safety hazard data. Based on the correlation test and correlation strength calculation, a safety hazard characteristic network is constructed, and the network and individual indices of safety hazard characteristics(out-degree, in-degree, intermediate, and eigenvector centralities) are analyzed to identify the key hazard characteristics(“facilities/equipment/apparatus/devices”, “settings”, “scaffolding”, and “railings”) and their associated paths with different roles. The early-warning strategies based on the data-driven network analysis are presented to improve timeliness from two aspects. In the early-warning process, relevant potential hazard information and importance ranking are obtained through the correlation analysis of the safety hazards via onsite inspection. Under limited manual labor, the safety management personnel are equipped with troubleshooting ideas and clues to mitigate the limitations of the original unplanned hazard inspection method. Meanwhile, a programmed information system can be further developed to provide early-warning tools. Safety managers can input the hazard information, and the system will conduct rapid standardization in the background to deal with the information and provide the associated hazards and sequencing to ensure timely feedback of hazard inspection clues. [Conclusions] This study establishes the standardization and characteristic extraction method of hazard record data which reduces data dimensionality from 3 267 nonstandard hazard records to 111 hazard characteristics and 11 hazard characteristic groups to clarify the key information regarding inspection and governance, such as hazard types and scenarios. Based on the analysis of individual and structural indices of the hazard characteristic network, the deduction and importance ranking of potentially associated hazard characteristics are realized, thus providing early-warning clues for effective safety inspection and governance. The early-warning strategies based on data-driven hazard characteristics can not only address the inefficient original unscheduled search method in the early-warning process but also improve safety management efficiency through timely feedback of inspection clues. This study introduces a foundational method for mining regular strategy information with early-warning clues of engineering construction safety hazard data, as well as for effective inspection and governance of construction safety hazards.