基于机器学习的锚固岩质边坡变形预测

锚索锚固是一种广泛使用的边坡加固技术，锚固性能的研究是锚固的核心问题之一。利用有限差分程序建立324组物理力学参数不同的锚固边坡，组成包括锚索参数和岩土体性质参数的9维输入指标和以沉降位移和塑性区面积为输出指标的数据集，分析输入输出指标间的关系。随后用随机森林和神经网络方法学习数据并建立层状边坡变形预测模型。分析显示，边坡沉降位移和塑性区面积预测结果变化对锚索性质参数中锚索总长度变化最敏感，锚固力的变化影响最小；其次岩土体物理力学性质中边坡力学指标黏聚力、内摩擦角起主要影响作用，岩土体密度变化影响最小；对预测结果的误差分析表明随机森林变形预测模型预测准确性比BP神经网络变形预测模型高5%~10%；模型预测沉降的偏差率小于预测塑性区面积的偏差率。研究表明随机森林算法在锚固效果预测问题上更加具有适用性，通过建立预测模型可以快速预测锚固边坡沉降位移和塑性区面积，指导锚固方案优化和变形控制设计。

Prediction of Deformation for Anchored Rock Slope Based on Machine Learning

Anchor cable anchorage is a widely used slope reinforcement technology and the research of anchorage performance is one of the core issues of anchorage. 324 groups of anchored slopes with different properties are established by using finite difference program. A 9-dimensional input index including anchor cable parameters and rock and soil property parameters and a set with settlement displacement and plastic zone area as output indexes are built. The relationship between input and output indexes is analyzed. Then, the random forest and neural network methods are used to learn the data, and the deformation prediction models of settlement displacement and plastic area are established. The results show that the settlement displacement and the area of plastic zone are more sensitive to the change of the property parameters of the anchor cable itself. The change of the total length of the anchor cable has the greatest influence on the output indexes while the change of the anchoring force has the least; Secondly, in the physical and mechanical properties of rock and soil mass, cohesion and internal friction angle exert the greatest influence and the change of rock and soil density has the least influence. The error analysis of the prediction results shows that the prediction accuracy of the random forest deformation prediction model is 5% ~ 10% higher than that of the BP neural network deformation prediction model. The deviation rate of the settlement predicted by the model is less than that of the plastic zone area. The research shows that the random forest algorithm is more applicable in the prediction of anchorage effect. By establishing the prediction model, the settlement displacement and plastic area of anchorage slope can be quickly predicted, which can guide the optimization of anchorage scheme and deformation control design.