基坑围护结构水平移动是其周围地表沉降的主要诱因之一。基于不同围护结构水平变形模式,根据线弹性理论相关研究给出了对应的地表沉降计算式。通过该计算式预测的黏土层中地表沉降最大值位置x_m与实测数据较为吻合。首先采用该计算式求出地表沉降最大值位置;其次,联合地层损失法,基于假设地表沉降曲线,计算沉降影响范围x₀,推导地表沉降曲线包络面积A_v;最后,根据地表沉降面积A_v与围护结构侧移面积A_h之间的相关性,计算地表沉降最大值δ_max,从而实现墙后任意地表位置沉降的计算。通过工程实例,验证了该方法的工程适用性。研究成果为基坑开挖地表沉降预测提供了一套半理论半经验方法。

Horizontal displacement of retaining structure is the main cause of ground surface settlement in deep excavation. In this paper we propose a method of estimating the ground surface settlement. First of all we assume that the ground surface settlement curve is a Gaussian probability density function, and derived the analytical formula of ground settlement under different lateral deformation modes of retaining wall based on linear elastic theory. Subsequently we acquired the location <i>x</i>_m of maximum settlement (which is defined as the distance <i>x</i>_m from the location of maximum settlement to foundation pit edge) in clay stratum based on the analytical formula, and then calculated the settlement-influenced range <i>x</i>₀ based on the aforementioned assumed function of ground settlement in association with soil loss theory. Furthermore, we derived the area <i>A</i>_v enveloped by the ground settlement curve, and in the meantime obtained the maximum ground settlement <i>δ</i>_max according to the relation between <i>A</i>_v and <i>A</i>_h, which is the area enveloped by the lateral displacement curve of retaining wall. The ground settlement of arbitrary location behind retaining wall hence can be estimated by substituting <i>δ</i>_max in the aforementioned assumed function. Through several engineering cases we proved that the proposed method is applicable, and thus providing a semi-theoretical and semi-empirical approach to predicting ground surface settlement.

为实现基坑变形预测及稳定性的综合研究,先以极限学习机(ELM)神经网络和灰色模型为基础,建立了基坑变形的串联、并联和混联耦合预测模型,以实现基坑变形预测;其次,再利用尖点突变理论和Mann-Kendall检验对基坑稳定性及变形趋势进行综合判断,以佐证预测结果的准确性。实例检验结果表明:3种耦合模型均能不同程度地提高预测精度,且以混联式模型的预测稳定性最高,其次是并联式模型和串联式模型;同时,预测结果与尖点突变理论和Mann-Kendall检验的分析结果相符,验证了该预测思路的有效性和可行性。研究方法可为基坑的变形预测提供一种的新思路。

In an attempt to comprehensively research the deformation prediction and stability of foundation pit, the series model, parallel model and parallel-serial coupled prediction model of foundation pit deformation are established on the basis of limit learning machine (ELM neural network) and grey model. Furthermore, the cusp catastrophe theory and Mann-Kendall test are employed to predict the stability and deformation trend of foundation pit to verify the correctness of prediction results. Case study show that the series model, parallel model and parallel-serial coupled model could all enhance prediction accuracy, among which the parallel-serial coupled model is of the highest stability, followed by parallel model and then serial model. In addition, the prediction results are in consistency with those by cusp catastrophe theory and Mann-Kendall test, indicating the effectiveness and feasibility of the present prediction method.

为准确掌握软土地区基坑侧位移变形特性,构建了基坑侧位移的预警模型和预测模型,其中,预警模型先以多重分形去趋势波动分析方法构建预警判别指标,再利用Spearman秩次检验实现判别指标的变化趋势判断,进而完成预警等级划分;预测模型则以脊波神经网络为基础,通过粗集理论和试错法优化模型参数,构建出优化变形预测模型。实例研究表明:通过预警分析,得出所给实例的预警等级为2级,说明其基坑侧位移趋于不利方向发展,应加强监测频率,提高施工安全预警;同时,在变形预测方面,参数优化能有效提高脊波神经网络的预测精度和稳健性,所得预测结果的平均相对误差均&lt;2%,具有较高预测精度,且其预测结果与预警结果一致,佐证了分析结果的准确性,可为基坑安全施工提供一定指导。

Early-warning model and prediction model for the side displacement of foundation pit were built in the aim of accurately grasping the deformation characteristics of foundation pit's side displacement in soft soil area. In the early-warning model, the early-warning discrimination indices were constructed using the multifractal detrended fluctuation analysis method, and then the change trends of the discrimination indices were determined by the Spearman rank test, hence the early-warning classification was completed. In the prediction model that is based on ridgelet neural network, the model parameters were optimized by rough set theory and trial and error method. Case study demonstrated that the early warning of the case in this paper was at level two, which indicated that the side displacement of the foundation pit tended to develop toward an unfavorable direction. Monitoring should be strengthened to improve the early warning for construction safety. In addition, the prediction accuracy and robustness of the ridgelet neural network can be effectively enhanced by parameter optimization, with the average relative error of the prediction results not exceeding 2%. The prediction results were consistent with the early warning results, which proved the accuracy of the analysis results.