尾矿坝溃决演进机制及监测预警方法研究

Tailings dam-break with limited emergency response time and huge potential threats, can often lead to heavy casualties and serious property losses. There are large numbers of tailings ponds located upstream of residential areas and important facilities in China, which are operating with an outdated emergency management mechanism. Therefore, the research on dam-break routing mechanism, advanced monitoring and early-warning methods is urgently needed. In this project, tailings dam safety will become the carrier of various disciplines and thus multidisciplinary technologies will be integrated and coordinated. A series of laboratory experiments will be carried out to study on the essential parameters, statics and dynamics properties of tailings in various subareas. Satellite remote sensing and photogrammetry will be applied into the interdisciplinary research on the theory and method of 3-dimensional monitoring system construction based on multisource data fusion. And according to the topographic 3-dimensional reconstruction results, combined with tailings spatial distribution law, the meshless particle dam-break model and reduced scale simulation experimental platform will be built. Furthermore, based on numerical simulation, similarity simulation and centrifuge experiment methods, the tailings dam routing laws and mechanism will be deeply revealed. And finally by training magnanimous data and disaster simulation visualization, the early-warning model based on artificial intelligence will be built, and emergency response grades will be proposed. This project has an important theoretical and practical significance for the improvement of assurance capacity and emergency management mechanism.

尾矿坝溃决应急响应时间短、潜在威胁巨大，往往造成惨重人员伤亡与巨额财产损失。我国“头顶库”数量多、应急管理机制相对落后，对于溃坝演进机制以及监测预警方法研究刻不容缓。本项目以尾矿坝安全为载体汇聚不同学科，融合多学科技术手段。以实验室试验系统研究各分区尾矿基本参数及静动力学特性；以卫星遥感、摄影测量技术研究多源数据融合的立体化监测体系构建理论与方法；以库区三维重建与尾矿分布规律为依据，建立溃坝无网格粒子模型，堆筑缩尺模拟试验平台；以数值仿真、相似模拟与离心试验相结合的方法，深入揭示溃坝演进规律及机制；最终，基于海量数据样本训练与灾害仿真可视化，创建基于人工智能的安全预警模型，划分应急响应等级，为提高尾矿库安全保障能力、完善应急管理机制提供理论依据与实现方法。

尾矿库溃坝灾害应急响应时间短、潜在威胁大，往往造成惨重的人员伤亡与巨额财产损失。为提高尾矿库安全管理水平，本项目在对国内外尾矿库进行系统调研分析的基础上，选取六座典型尾矿库为研究对象，结合现场调查与测试、实验室试验、数值模拟仿真与理论分析方法，对尾矿库无人机监测、尾矿坝溃坝演进仿真、动静载荷下尾矿坝稳定性、尾矿库浸润线预测及预警管理等内容，展开了系统的基础理论与应用方法研究，主要包括以下内容：1）基于无人机倾斜序列影像的尾矿坝边坡表面变形监测方法；2）基于机器学习的尾矿库图像特征提取与匹配方法；3）高浸润线条件下的尾矿坝渗流破坏模式及过程的常规及离心试验模拟；4)高浸润线条件下尾矿坝稳定性数值模拟分析；5）基于光滑无网格粒子法（SPH）的溃坝泥浆演进过程仿真；6）基于深度学习的尾矿库浸润线预测方法及尾矿库安全预警管理系统开发。.课题研究主要取得以下成果：1）提出了基于深度学习的图像特征提取与匹配算法，有效提升了尾矿库图像拼接速度与精度，提高了工程现场的无人机倾斜序列图像采集处理质量；2）提出了基于无人机倾斜序列影像的尾矿坝边坡表面变形监测方法，精度达到厘米级，实现大范围、全覆盖快速监测，为尾矿库现代化安全监测提供了有力的技术支持；3）通过高浸润线条件下的尾矿坝常规与离心试验，揭示出常规尾矿坝渗流破坏形式及其与坝体材料性质的关系；4）基于实验室缩尺物理模拟与溃坝事故案例数值仿真验证，构建出尾矿库溃灾演进SPH计算模型，为尾矿库溃坝灾害影响范围确定及防治方法选择提供了依据；5）通过深度学习的GRU、LSTM、RNN、Dense四种模型加上机器学习的SVM、GBR两种模型组合分析表明，GRU应用于浸润线预测具有优越性和可行性；6）基于Python语言开发设计了尾矿库安全管理预警系统，不仅具有预警软件的相关功能，还额外增加了浸润线预测功能，做到了防患于未然。