基于BP神经网络的重塑黄土冻融过程抗剪强度劣化特性研究

Shear Strength Deterioration of Remolded Loess During Freezing-thawing Process Based on BP Neural Network

冻融循环导致黄土强度的损伤规律十分复杂,传统单一因素评价方法难以确定冻融过程黄土抗剪强度指标损伤规律与多因素之间的量化统计关系。基于此,首先针对西安Q3重塑黄土进行室内冻融试验,得到不同干密度、含水率及冻融次数下的抗剪强度指标数据;然后采用BP神经网络算法对试验数据进行学习训练,得到各因素与抗剪强度指标间的预测模型。研究发现:黄土试样黏聚力随冻融次数增加呈指数衰减趋势;黏聚力随含水率和干密度增加分别表现出线性衰减和增加特征且冻融后黏聚力与含水率和干密度的变化曲线近似重合;内摩擦角呈波浪形变化趋势且波动范围较小,无明显变化。冻融过程黄土黏聚力神经网络模型的预测值和试验值之间相对误差较小,表明该方法具有较好的预测精度,能够综合描述诸因素与黏聚力的量化关系。

The damage law of loess strength caused by freezing and thawing cycles is very complicated. However,the traditional method based on only one factor is difficult to quantitatively determine the statistical relationship between the damage law of shear strength and many factors. Based on that,freeze-thaw test was firstly carried out to get the shear strength index data of Xi＇an Q3 remolded loess under different dry density,moisture content and freezethaw times. Then a prediction model was developed using testing data based on BP neural network. The results show that the cohesion of loess decreases exponentially with freezing and thawing times. With the increase of water content and dry density,the cohesion decreases and increases linearly respectively,and has approximately same variation after freezing and thawing. The internal friction angle changes undulate with a small amplitude and has no obvious variation during the whole freezing-thawing process. The relative error of prediction value of cohesion compared with experimental data is little,indicating that BP neutral network forecasting method has good accuracy and can describe the quantitative relationship between factors and cohesion.