基于大数据理论的土石坝安全态势感知模型和方法研究

Since complex geological conditions,uncertain external load, structure anisotropy of embankment dam,the factors that affect the safety Situation have strong spatial variability and time variation. Its Safety Pre-Warning needs to describe the dynamic process of the safety situation and to predict the safety tendency, which means the work of embankment dam safety situational awareness.With the development of surveillance and communication technology, the embankment dam monitoring data showing an increasingly characteristic of large data,such as volume,diversity, high value,variability .Based on the large data it constructs a reliability function and establishs the analytical model of unstructured dam safety monitoring data. With cloud theory unstructured data and structured data has been fused. Based on the fusion rule, the factors that influence the safety of the embankment dam will be mined according to the database of the failure dam and dangerous weak reservoir by sorting algorithms. And the weight of the influence factor will be determined by Recursion Artificial Neural Network. The multi-scale characteristics of cloud fusion data will be extracted by scale entropy, the relationship among the monitoring parameters will also be revealed, and thus the three dimensional (Real Time Monitor-Operations Management-Safety Situation) time-space evolving model will be built up.The characteristics of dam Effect-Quantity treate as information granulation on this basis.Using the method of support vector machine,it can creat the model of embankment dam safety situational awareness,and be verified by the case of typical dams.This research has important theoretical and practical significance for improving and enriching dam safety monitoring theory and methods, and guiding disaster prevention and mitigation work of water conservancy under disaster environment.

土石坝地质条件复杂、外部荷载不确定、自身结构各向异性，影响其安全性态的因素具有较强的空间差异性及时变性，其安全预警需要表征安全性态的动态演变过程和预测安全趋势，即对土石坝安全态势进行感知。随着监测和通信技术的发展，感知土石坝安全态势的监测数据日益呈现出体量大、多样性、价值高、变化性的大数据特征。本项目拟基于大数据理论，通过云变换，实现监测数据非结构化和结构化云融合；在此融合规则下，对溃坝和病险水库数据库采用分类算法挖掘土石坝病害因子，并利用递归神经网络进行因子赋权；采用尺度熵提取云融合数据的多尺度特征，挖掘各监测量间关联规则，建立“实时监测维度、运行管理维度、安全性态维度”三维时空演化模型；对监测量各维尺度特征进行信息粒化处理，结合各因子权重，利用支持向量机训练，建立土石坝安全态势感知模型，并以工程实例加以验证。该项研究对提高土石坝安全预警水平，保障大坝长效服役，具有重要的理论和现实意义。

土石坝地质条件复杂、外部荷载不确定、自身结构各向异性，影响其安全性态的因素具有较强的空间差异性及时变性，其安全预警需要表征安全性态的动态演变过程和预测安全趋势，即对土石坝安全态势进行感知。随着监测和通信技术的发展，感知土石坝安全态势的监测数据日益呈现出体量大、多样性、价值高、变化性的大数据特征。本项目基于大数据理论，通过云变换，实现了监测数据非结构化和结构化云融合；在此融合规则下，对溃坝和病险水库数据库采用分类算法挖掘土石坝病害因子，利用递归神经网络进行因子赋权；提出了安全监测数据标注方法，建立了大坝工程缺陷和安全隐患的监督学习模型，支持大坝安全的决策边界；提出了大坝安全四级预警指标拟定方法；划分了基于监测数据、现场安全检查、安全复核结果的风险等级辨识标准。建立了多源多模态数据特征提取模型，挖掘各监测量间关联规则，建立了“实时监测维度、运行管理维度、安全性态维度”三维时空演化模型；结合各因子权重，利用支持向量机训练，建立了土石坝安全态势感知模型，并以澜沧江苗尾水库工程实例加以验证。利用MYSQL开源数据库进行构建大坝安全态势感知知识库，开发了广西自治区水库大坝安全智慧管理大数据服务平台，对工程实际应用。该项研究成果为我国水库大坝安全智慧决策提供了技术支撑，对提高土石坝安全预警水平，保障大坝长效服役，具有重要的理论和现实意义。