面向大规模复杂数据的地铁施工安全多粒度知识发现与动态风险感知研究

Metro construction safety violations are affected by various factors, such as geological and hydrological conditions, construction methods, management levels and so on, resulting in challenges in identification of cataclysmic laws on metro construction accidents. Currently, the empirical knowledge of field experts is greatly relied in construction safety early-warning practice. With the rapid development of construction informatization and network techniques, the conflict between rich data and poor knowledge attracts increasing public attentions. Traditional physically handling approaches are difficult to deal with large-scale and complex data, leading to the reduce of accuracy and effectiveness regarding the safety risk perception in metro construction . In this project, both granular computing and knowledge discovery are utilized to explore the methodology of information granulation in metro construction, in order to improve the cognition levels of large-scale and complex data in terms of information granulation structure. Combining granular computing and data mining techniques together, this project then investigates the multi-granularity knowledge discovery approach based on large-scale and complex data, acquire cataclysmic rules from huge data sets, and further reveals the evolution laws with regard to metro construction accident. On this basis, a dynamic risk perception approach based on dynamic Bayesian network in metro construction is finally proposed, with both the feed-forward and feed-back risk analysis and assessment taken into account, which can provide real-time decision support on risk control and emergency rescue in metro construction.

地铁工程施工安全事故发生受地质、水文、工法及管理水平等多种因素共同作用，灾变演化规律复杂，现场施工安全预警对领域专家经验知识依赖程度大。随着工程信息化及网络技术的日益普及，丰富的数据与贫乏的知识之间的矛盾日见突出，传统人工处理方式对大规模复杂数据的处理与分析存在较大不足，影响了地铁施工安全风险感知与决策控制的准确性与实效性。本项目以粒计算和知识发现为理论基础，研究适用于地铁工程大规模复杂数据的信息粒化处理方法，提高对大规模复杂数据知识粒度结构的认知水平；将粒计算与数据挖掘技术有机结合，探索面向大规模复杂数据的多粒度知识获取与发现方法，从海量数据中获取地铁工程施工安全事故的灾变路径与机制，进一步揭示地铁工程施工安全灾变的演化规律；在此基础上提出基于动态贝叶斯网络的地铁工程施工安全动态风险感知方法，实现安全风险的实时前馈与反馈预警评价分析，为地铁工程施工安全风险控制与应急救援提供动态决策支持。

以研究盾构地铁结构运营安全影响规律为出发点，以致险因素影响规律分析-数据采集与处理-突发荷载下地铁运营安全时变评价-运营盾构地铁的安全可靠性全面精准评估为主线，结合采用有限元建模、代理模型拟合、敏感性分析、结构健康监测、压缩感知技术、Copula理论、贝叶斯更新理论、可靠性分析理论、MCMC算法、Pair-Copula分解理论和贝叶斯网络等技术，深入分析周围多源不确定性因素作用对盾构地铁结构安全性的影响规律，利用健康监测技术获得盾构地铁结构实时状态监测信息，并利用贝叶斯更新技术、贝叶斯网络模型和Copula理论构建相依性PCBN模型对运营地铁的安全状态进行精确评估，从而为运营盾构结构的安全管理提供信息支持