摘  要

地铁施工是一项高风险建设项目，由于下穿城市腹地，周边环境复杂，地质条件等不确定性因素较多，组织管理难度大，安全事故频发，因此迫切需要有效的安全风险分析模型和风险管理手段来预防事故的发生。然而，地铁施工过程是一个复杂社会技术系统，传统方法缺少针对地铁施工特点进行系统性的风险建模，导致分析结果往往较为片面，在实际工作中难以有效指导安全风险管理。本文尝试从系统论、控制论的角度构建针对地铁施工过程的系统安全风险分析模型，识别安全风险演化的客观规律，为预防地铁施工导致的重大事故提供决策依据。主要研究工作和创新性成果如下：

本文分析了构建系统安全风险分析模型的现状、动机和挑战，并结合地铁施工的特点总结了地铁施工安全风险建模需要解决的问题，明确了建模的目的、意义和边界，将地铁施工系统划分为组织、技术和环境三个子系统，分别代表地铁施工过程中的组织管理、施工过程和环境影响，提出了各子系统的建模目标和建模难点，并从系统论、控制论的角度提出了地铁施工安全控制结构，总结了各子系统之间的相互关系。

针对地铁施工技术子系统，为了建立几何参数、地质参数、盾构参数等影响因素与地表变形的复杂非线性关系，并优化盾构施工参数，本文基于平滑相关向量机、自适应高斯核函数和粒子群优化方法建立了地表沉降分析预测模型，具有预测精度高、能识别因素重要性和算法效率高的特点。在此基础上，通过对下一步盾构施工的参数进行优化，构建了盾构施工过程的前馈控制模型。

针对地铁施工环境子系统，为了使传统以专家判断为主、结合数据分析的风险分析过程显性化、自动化、可计算化，本文构建了一个具有四层层次结构的概率风险评估模型框架来描述风险的传播过程，利用历史数据和专家判断，通过混合相关向量分类机和贝叶斯网络建立了地铁盾构施工引起的周边建(构)筑物、管线和路面风险的定量评估模型，给出了地铁施工引起周边环境风险的事故场景、后果和概率，从而判断出地铁施工工点的风险等级。

针对地铁施工组织子系统，为了分析地铁施工中的组织管理因素与系统安全绩效之间的关系，本文构建了组织安全绩效模型框架，分析了框架下安全文化、安全管理系统、安全绩效的定义、维度以及系统变量之间的关系，探讨了组织管理因素是如何决定安全管理实践并最终影响系统安全绩效的，然后分析了安全绩效与成本和进度绩效的关系，强调了安全管理承诺对于平衡安全目标与生产目标之间的重要性。在此基础上，以系统动力学作为建模工具分别对组织安全、成本和进度三个绩效目标进行了定性建模，描述了面向安全的组织学习、决策和管理是如何进行的。

最后，在分析了技术、环境和组织子系统之间的相互作用关系的基础上，通过整合各个子模型，建立了地铁施工安全风险动态演化模型(Dynamic Risk Assessment For Tunnelling system，DRAFTs)，用于分析地铁施工过程中的风险演化规律和事故发生原因。通过仿真，分析了地铁施工过程中的进度、成本和安全绩效以及安全风险的时空演化规律，描述了组织管理因素和技术因素是如何相互作用并导致事故的发生，为预防地铁施工过程中类似事故的发生提供了新的解决思路。

关键词：风险演化；地铁隧道施工；(平滑)相关向量机；贝叶斯网络；系统动力学

Abstract

There is an intrinsic risk associated with urban tunnelconstruction because of the limited a priori knowledge of geotechnicaluncertainties, the potential for

causing damages to adjacent structures,facilities and pipelines and the complex organizational management processes. Toprevent the tunneling-induced major

accidents from occurring, it is necessaryto implement a reasonable safety risk analysis during tunneling. Tunnelingprocess is a complex socio-techincal system

. However, most current riskanalysis models lack a systematic view and the very nature of tunnelconstruction is not completely considered, resulting in limited applications. Thisthesis attempts to establish a safety risk model based on system theory andcontrol theory to identify the accident causality underlying thetunneling-induced major accidents and thus provide decision making support. Themain contents and highlights of the thesis are summarized as follows:

This thesis first analyzes the current trends, stimuliand challenges of establishing a systemic safety risk model. By considering thevery nature of tunnel

construction, the safety problems during tunneling areidentified. Then, the goals, assumptions and boundaries of the model aredetermined. Based on the modeling assumptions, the tunneling system is dividedinto three sub-systems, i.e., organizational, technical and environmentalsubsystem, which represents the organizational

managemen process,  the tunnelconstruction process and the tunneling-induced damages to the environment respectively.The modelling aims and challenges

of each sub-system are also clarified. Moreover,the basic safety control structure of tunnel construction is proposed as arepresentation of the

interrelationship among the sub-systems.

The aim of technical subsystem modelling is toextract the complex nonlinear relationship between the geometrical, geological,shield operation parameters and

the ground surface settlement during the constructionprocess, and optimize relevant construction parameters. This thesis establishesa settlement prediction model based on smooth relevance vector machine withadaptive Gaussian kernel function and particle swarm optimization. The applicationof this method indicates that the model is accurate and effective, and canidentify the relative importance of each factor. Based on the establishedrelationship, the shield operational parameters for next

excavation step canthen be optimized by minimizing both the induced settlement and adjustment ofthe shield parameter, thus forming a feedforward control process.

         The aim of environmental sub-system modelling is torepresent the risk analysis that relies on expert judgment and data analysis ina explicit and structured way.

A four-level hierarchical framework is proposedto describle the risk propagation. By combining relevance vector classifier and Bayesian network under the

proposed framework, the accident scenarios,consequences and probabilities of damages to structures, pipelines and groundsurface are qualitatively and quantitatively depicted using both historicaldata and expert judgments. The risk level of each tunnel section is therefore determinedby comparing the analysis results and the risk acceptance criteria.

The aim of organizational sub-system modelling is todescribe the relationships between the organizationl factors and the systemsafety performance. The

definitions, dimensions and inter relationships ofsafety culture, safety management system and safety performance are reviewed,which gives insights into how the

organizational factors influence the safetymanagement practice and the subsequent organizational safety performance. Theinteraction between safety pressure and

schedule/cost pressure is alsodiscussed. The management commitment towards safety is found very important inbalancing the production and protection. Based on

the analysis, organizationalperformance models under safety, cost and schedule objectives are qualitativelypresented using system dynamics, which describe how

the organizationallearning, decision-making and management are taking place.

       Finally, by integrating the three sub-systems, a DynamicRisk Assessment For Tunnelling system (DRAFTs) is proposed to analyze the riskdynamics and

accident causality during tunneling. The simulation results showthe cost, schedule and safety performance variation during tunnel constructionand how the

organizational factors influence the technical failure andcontribute to the accident, which gives foresight to prevention of majoraccidents during tunneling.

Key words：risk dynamics; tunneling; (smooth) relevance vectormachine; Bayesian belief network; system dynamics