大数据环境下的土石坝病险挖掘和诊断

China has built more than ninety thousand dams, most of them are earth rockfill dams. The dangerous problems of earth rockfill dam have endanger the dam safe operation. The existing research focuses on a single project or a dangerous problem. With the construction of dams and dam foundation centralized regulatory information platform, there has been a large number of earth rockfill dam comprehensive big data information. Based on big data, the common dangerous mining approach of earth rockfill dam can reveal the rule of dangerous information which hiddens in a lot of accident dam samples. Therefore, combined with the big data analysis and the theory of dam safety analysis, the research on dangerous mining and diagnosis of earth rockfill dam in the big data environment is proposed. Through the extraction of dam dangerous mode, the dam diagnosis evolution path and network is established which has quantitative relationships between implicit, uncertain, and deterministic. Based on the study of the sensitivity and correlation mining methods of the dangerous, the dam dangerous factors are identified. The multi-path optimization algorithm based on PSO is proposed to achieve recognition of main evolution path. With application of Bayesian theory, for a specific dangerous dam, the Bayesian dynamic diagnosis model is established to achieve the dam main factors and evolution path. The research can provide guidance for dam safety management and dangerous analysis, and has important theoretical significance and scientific value.

国内9万余座大坝中大部分为土石坝，土石坝的病险问题一直危及到大坝的安全运行。现有的研究多关注单个工程或一种病险问题展开。随着水库大坝的集中监管及水库大坝基础信息平台的建设，产生了大量土石坝病险相关的全方位大数据信息。从大数据的角度挖掘土石坝病险共性，能够发觉大量病险土石坝样本信息下隐藏的病险规律。为此，将大数据分析方法和土石坝安全分析理论结合，研究大数据环境下的土石坝病险挖掘和诊断，包括对土石坝病险模式进行提取，构建包含隐性、不确定性和确定性量化关系的土石坝病险演化路径和网络；研究病险诱因的相关性和敏感性挖掘方法，实现土石坝病险因素识别；研究基于粒子群的多路径寻优算法，实现土石坝病险演化主要路径的识别；应用贝叶斯理论，针对特定病险坝，建立土石坝病险的贝叶斯动态诊断模型，实现土石坝病险因素及演化主要路径的诊断。研究可为土石坝安全运管、病险分析提供指导，具有重要的理论意义和应用价值。

本项目通过病险工程调研、有限元计算分析、物探、查阅水库大坝病险数据库和资料文献，来收集土石坝病险信息。首先，开展土石坝病险演化关系的研究，构建能够表征土石坝病险规律的病险演化路径和网络关系；其次，开展土石坝病险信息的数据挖掘方法研究，实现土石坝病险演化网络中病险诱发因素和主要病险演化路径的分析；最后，研究土石坝病险演化网络的转化和更新方法，建立土石坝病险的贝叶斯动态诊断模型。整个研究过程，以建立土石坝病险演化路径为基础，将土石坝病险信息数据挖掘研究作为核心，以面向土石坝病险实时诊断分析为目的，形成三个逐层递进的研究过程。研究为识别水库大坝病险、评估大坝致灾风险、有效防灾减灾、提高灾后应对处置能力提供支持，从而保障土石坝安全运行。