基于动态贝叶斯网络与随机有限单元法的尾矿库安全可靠性评价

The accurate evaluation of tailings dam stability is an important basis for ensuring the safe operation and construction of tailings dam. However, the factors affecting the stability of the dam, such as the particle distribution of tailings, the height of wetting line, and the mechanical properties of tailings, have complex interactions and these interactions cannot be expressed quantitatively by mathematical functions. Therefore, the uncertainty of the basic data of dam stability assessment will significantly reduce the accuracy of the evaluation results. Based on many assessment data to tailings dam stability, the project will carry out similar simulation experiments to investigate the correlations among factors above. Then, a dynamic Bayesian network (DBN) model based on Bayesian theory is established to optimize the basic data of the stability analysis of the tailings dam. And a machine learning method is proposed to calculate the conditional probability relationship among each node variables in DBN model. Output data of DBN work as the input data of the newly developed model based on stochastic finite element method for the calculation of the dam stability. Finally, an early warning system of tailings dam instability would be established. This study not only enrich the interrelationship between the factors affecting the instability of tailings dam but also develop the slope safety evaluation algorithm model, which has important theoretical and practical implications for the safe operation of the tailings dam and slope stability assessment techniques.

尾矿坝稳定性的精确评价是确保尾矿库安全修筑运营的重要依据。然而，影响坝体稳定性的各因素，如尾矿库颗粒分布、浸润线高度、尾矿力学性质等，具有复杂的相互作用，且这些作用难以用具体的数学函数关系来定量表达，以致尾矿坝稳定性评价基础数据的不确定性，将显著降低评价结果的准确性。为此，本项目拟在收集大量尾矿坝稳定性评价数据基础上，开展影响尾矿坝各因素间关联性相似模拟试验。而后，基于贝叶斯理论建立用于优化尾矿库稳定性分析基础数据的动态贝叶斯网络（DBN）模型，并提出实现DBN模型中各节点变量间的条件概率关系的机器学习方法。同时，以DBN输出结果为依据，基于随机有限单元法，建立一种尾矿库稳定性评价强度折减模型，并构建基于该模型与评价方法的尾矿库早期失稳的预警系统。本研究成果不仅丰富了尾矿库失稳影响因子的相互关系，且发展了边坡安全性评价算法模型，对于促进尾矿库安全运营，发展边坡稳定性评价技术均具有重要的理论

中国作为一个矿业大国，现有尾矿库 8000 余座，居世界第一。尾矿坝的安全是矿山安全研究中非常重要的一部分，而尾矿坝稳定性研究是其中最主要的研究方向之一。然而，影响坝体稳定性的各因素不确定性高、关键因子的预测准确度低、尾矿坝失稳破坏机理不明确等极大影响了尾矿坝安全性研究。本项目首先对边坡现场进行取样，建立室外大型相似物理模型试验和室内尾矿坝溃坝实验，通过实验获取了边坡溃坝机制，依据实验结果对边坡稳定性因子进行了多因素动态分析；而后，通过建立动态贝叶斯理论模型降低了尾矿边坡稳定性影响因子的不确定性，实现了对尾矿坝浸润线的预测和估计；然后提出了更加准确评价尾矿边坡稳定性的随机理论模型；最后利用研究成果在边坡现场建立了安全监测预警系统。此项目可为尾矿坝溃坝行为预测和防灾减灾提供理论和实践依据。