本体驱动下的建设工程安全事故网络模型构建、解析与应用研究

Construction accidents with high frequency have a negative impact on individual, organization and nation. They definitely result in tremendous economic loss and immediate cost. It is significant to study how to reduce the possibility of accident occurrence and attenuate its influence. On the basis of ontology, this study will conduct accident network modeling to innovate the method for construction safety management. The energy release theory will be adopted to analyze the complicated process of construction accidents in terms of energy transference. The advantage of the energy release theory is that it reveals the nature of accident occurrence from the ontological and physical perspective. The method of METHONTOLOGY will be modified to be adaptive to building construction safety accident ontology (OntoCSA). Considering its simplicity and ease of use, WebProtégé will be adopted as the tool for editing OntoCSA. In addition, its functionality can be extended using a plug-in architecture and a Java-based application programming interface (API) for implementation. As to the construction safety accident network (CSAN) modeling, heterogeneous nodes and arcs need identifying. The weights of heterogeneous arcs will be determined by co-occurrence analysis method. Moreover, the complex network theory is an available approach to exploring the complexity of CSAN and analyze its topological characteristics from macro-dimension. There are two main parts of the quantification of construction safety risk through the hybrid Bayesian network inference. The first part aims to determine the most possible causing factors according to a consequence incident, and the second part aims to predict the probabilities of corresponding derivative incidents. Considering the discrepancy of node-trigger modes (including single-trigger and multiple-trigger mode), and variable types of elementary event probability (including 0-1 discrete and continuous variable), the optimization models can be built based on inter linear programing and polynomial nonlinear programming, respectively, in order to determine the best intervention strategy for the reduction of construction safety risk. The principal contribution of this research is intended to be the generalization of these concepts and models and the application of optimization methods to the related decision-making processes in the area of construction safety management.

持续不断的建设工程安全事故给个人、组织和国家，带来巨大经济损失与间接成本，因此研究减少建设工程安全事故发生的可能性，降低事故影响的严重性，提高其安全绩效具有重要现实意义。本项目基于本体驱动，面向事故网络建模，创新建设工程安全管理方法。首先从事故本体角度构建建设工程安全事故的能量意外释放模型，并改进METHONTOLOGY方法，应用WebProtégé，构建建设工程安全事故本体。在此基础上识别CSAN异质节点与边，采用共现分析法确定边权重，构建有向加权异质的CSAN。进而，运用复杂网络和混合贝叶斯网络解析CSAN，定量化计算建设工程安全事故风险。最终，针对CSAN节点触发模式的差异（单触发与多触发模式），以及基本事件发生概率的差异（0-1离散型变量与连续型变量），分别基于整数线性规划与多项式非线性规划构建面向CSAN风险降低策略的最优化模型，为解决建设工程安全管理问题提供理论支撑与实践指导。

首先，本课题基于建设工程安全事故数据运用能量意外释放理论，将事故致因分为人、物、环、管四类，依据国家标准或规定，形成一套规范化的事故致因和事故类型分类标准，作为事故案例分析依据。运用网络理论探究事故的发生机理和内在成因机制，提取事故案例中的因素形成事故链，并使用图结构相似性度量的方法验证了事故致因和事故类型分析标准的效果。进而，将事故链整合成建设工程安全事故致因网络，完成网络模型构建。. 其次，运用网络拓扑特征量从宏观上分析网络模型的拓扑结构。对选取的度分布、聚集系数、平均最短路径、介数中心性等多个参数进行计算与分析，表征地铁建设工程安全事故致因网络无标度网络特性和小世界特性。并通过随机目标攻击和特定目标攻击分析建设工程安全事故网络的鲁棒性和脆弱性，研究节点遭受攻击后的级联失效作用，确定对网络造成的扰动结果，找出对事故预防起到关键作用的关键节点。. 再次，运用贝叶斯网络对建设工程安全事故网络模型风险的定量化研究。在构建建设工程安全事故致因网络模型的基础上，通过贝叶斯网络的正向因果推理，预测不同致因因素或者原生事故下的工程建设安全事故风险的发生概率。其次利用反向诊断推理分析建设工程安全事故的成因机理，确定最可能致因因素或原生事故等，从多个不同角度对建设工程安全事故风险水平进行量化分析。. 最后，建立建设工程风险降低模型，探究最优安全策略。按照节点触发方式不同，构建不同的风险降低模型。针对高风险等级事件，进行实证研究。分别从风险降低值和投入效益的角度探讨最优安全策略。通过计算，当仅对单个敏感因素投入安全策略时，结构强度不够的风险降低值最高，但是冒险进入危险场所的投入效益最高；同时对多个敏感因素投入安全策略时，冒险进入危险场所+风险监控缺失或未及时响应的策略组合的投入效益最高。因此在工程建设过程中，可以针对这两种敏感因素投入安全策略，有效降低结构坍塌事件和物体打击事件的安全风险。