基于梯度提升决策树的露天矿边坡多源信息融合与稳定性预测

Multi-source information fusion and stability prediction of slope based on gradient boosting decision tree

准确预测边坡稳定性对于减少滑坡次数和降低边坡维护成本至关重要。岩质边坡作为一种典型的岩体工程,在灾害的孕育和发展过程中,工程岩体内部会产生新的裂隙,或旧裂隙发展。随着岩体内部裂隙的发展,岩体物理参数不断发生变化,因此通过监测边坡岩体的物理参数变化对岩质边坡稳定性进行预判已成为一种重要的预测手段。而传统的单一监测信息虽然能够直观地反映滑坡趋势,但其存在局部性和滞后性,并不能完整反映边坡所处的状态。基于涵盖了边坡内部和表面的微震、应力和位移等3种异构信息,提出一种利用多源监测信息融合技术对边坡进行稳定性预测分析的方法:针对大孤山露天矿边坡的实际地质条件,通过有限元强度折减法得到边坡不同状态下监测信息的变化规律,并利用梯度提升决策树(GBDT)模型对监测信息进行融合,建立了边坡稳定性预测的非线性模型,并与不同融合算法进行对比,得到如下结论:(1)将梯度提升决策树和有限元强度折减法相结合,可以实现边坡的位移、应力和微震等异构信息的融合,并以大孤山铁矿西北帮边坡实际监测数据验证了所提方法的有效性;(2)与其他融合算法进行比较表明,GBDT模型在预测精度和模型解释能力方面具有优越的性能,可以很好地识别复杂的非线性关系,适用于露天矿边坡安全系数的预测;(3)基于GBDT算法,可以计算不同监测信息的变化对边坡安全系数的影响,研究表明,坡顶的位移变化和边坡内部微震数对于边坡的稳定性预测占主导作用。

Accurate prediction of slope stability is very important to reduce the frequency of landslides and the cost of slope maintenance.For rock slope engineering,there will be new cracks or old cracks inside the engineering rock mass in the process of disaster development.With the development of the internal fracture of rock mass,the physical parameters of rock mass will change constantly,therefor it has become an important method to predict the stability of rock slope by monitoring the changes of physical parameters of rock mass.However,the traditional single monitoring information can directly reflect the landslide trend,but it has the characteristics of locality and delay,and cannot completely reflect the state of the slope.Based on the three heterogeneous information including micro-seismic,stress and displacement,this paper puts forward a method of using multi-source monitoring information fusion technology to predict and analyze the slope stability.According to the actual geological conditions of Dagushan open pit mine slope,the change rule of monitoring information in different states of slope is obtained by the finite element strength reduction method,the GBDT method is used to integrate the monitoring information,and the nonlinear model of slope stability prediction is established.The conclusions are as follows:(1)the combination of the GBDT method and the finite element strength reduction method can realize the fusion of heterogeneous information such as displacement,stress and micro-seismic of the slope,and the monitoring data in the actual slope of Dagushan Iron Mine is used to verify the effectiveness of the proposed method;(2)compared with other fusion algorithms,the GBDT model has superior performance in prediction accuracy and model interpretation ability.The model can identify complex nonlinear relations well,and is suitable for the prediction of slope safety in the open pit mine;(3)based on the GBDT algorithm,it can calculate the weight of different monitoring information for slope safety.The study shows that the displacement change of the slope top and the number of microseisms inside the slope play a leading role in the stability prediction of the slope.