论文部分内容阅读
采用剩余推力法与BP神经网络,以贵州省毕节地区宋阴公路K5+170~K5+220段玄武岩残坡积土边坡作为工程研究对象,对该边坡稳定性展开了计算和预测。选取现场实测剖面作为计算剖面,设置4个计算工况,由剩余推力法得到边坡天然状态(工况1)稳定性系数为1.085 1,当边坡处于16m地下水位+暴雨(工况2)、16m→8m地下变化水位(工况3)和16m→8m地下变化水位+暴雨(工况4)时,边坡稳定性系数均小于1。边坡稳定性敏感因素分析显示,滑带土黏聚力敏感系数平均值为15.9%,内摩擦角为48.3%,地下水位为34.0%,表明滑带土内摩擦角对边坡稳定影响最大,其次是地下水位。选择同一路段其他玄武岩残坡积土滑坡作为训练样本,通过Matlab神经网络ANN工具箱分步骤设计了BP网络,选择加动量学习速率自适应traingdx函数作为训练函数,采用多次预测求均值的方法获取预测结果。BP神经网络预测结果表明,边坡工况1的稳定性系数平均值为1.095~1.139,工况3为0.988~1.021,考虑到暴雨对边坡坡稳定性的影响,工况4时边坡可能发生滑动破坏。神经网络各次预测结果之间误差较大,最大达到45.87%,但求均值后的BP神经网络预测结果与剩余推力计算结果的相对误差大大降低,仅为0.4%~5.2%。将BP网络的输入参数减少为5个后,预测精度反而较高,表明黏聚力、内摩擦角、坡高、坡角、湿重度等因素对边坡稳定性有着实质性的影响,其他因素影响权重则较低。
Using remaining thrust method and BP neural network, the slope of basal residual rock slope of K5 + 170 ~ K5 + 220 Songyin Highway in Bijie Prefecture of Guizhou Province was taken as the engineering research object, and the stability of the slope was calculated and predicted. According to the remaining thrust method, the stability coefficient of the natural state of the slope (condition 1) is 1.085 1. When the slope is at 16m groundwater level + heavy rain (condition 2) , 16m → 8m underground water level (condition 3) and 16m → 8m underground water level + heavy rainfall (condition 4), the slope stability coefficients are all less than 1. Sensitivity analysis of slope stability shows that the average coefficient of cohesion of slide zone soil is 15.9%, the internal friction angle is 48.3%, and the groundwater level is 34.0%. It shows that the friction angle within the slide zone affects the slope stability the most, Followed by groundwater level. The other basalt residual landslides on the same road section are selected as training samples. BP neural network ANN toolbox is used to design the BP network step by step. The adaptive traingdx function with momentum learning rate is selected as the training function, and multiple prediction averaging methods are used forecast result. BP neural network prediction results show that the average stability factor of slope condition 1 is 1.095 ~ 1.139 and the condition 3 is 0.988 ~ 1.021. Considering the influence of heavy rain on the slope stability, Sliding damage occurred. The error between each prediction result of neural network is larger, reaching a maximum of 45.87%. However, the relative error between the prediction results of BP neural network and the remaining thrust calculation results is greatly reduced, only 0.4% -5.2%. After the input parameter of BP network is reduced to five, the prediction accuracy is rather high, indicating that cohesion, internal friction angle, slope height, slope angle, wet weight and other factors have a substantial impact on slope stability. Other factors Impact weight is lower.