路基边坡降雨试验及基于神经网络的水分场研究  

Rainfall infiltration test on subgrade slope and study on moisture migration by neural network

作　　者：吴谦[1] 王常明[1] 王天佐[1] 黄晓虎[1] 张志敏[1] 张兆楠 

机构地区：[1]吉林大学建设工程学院,吉林长春130026

出　　处：《中南大学学报（自然科学版）》2016年第4期1237-1244,共8页Journal of Central South University:Science and Technology

基　　金：国家自然科学基金资助项目(40972171);吉林大学研究生创新基金资助项目(2014065)~~

摘　　要：为了研究降雨条件下路基边坡土体含水率的变化规律及雨水入渗过程,进行一系列不同降雨强度、不同坡度条件下的室内降雨模拟试验。以土体含水率受降雨历时、土体空间位置、坡度和降雨强度这4个因素作为输入单元,体积含水率作为输出单元,选取试验数据输入训练,建立含水率的遗传算法神经网络预测模型。预测检验后,利用神经网络对2.7 mm/min降雨强度下40°边坡的降雨入渗过程进行预测研究。研究结果表明:对于路基边坡,当土的性质、压实度、排水等条件相同时,土体含水率受降雨历时、土体空间位置、坡度和降雨强度这4个因素共同影响;随降雨的进行,土体含水率逐渐增加,浸润范围不断增大,受空间位置影响距入渗面越远则含水率变化滞后,增长速率及幅度减小;在相同雨强下,不同坡度边坡坡顶土体含水率变化过程相似,而随坡度的增大,坡脚土体含水率的增长速率及幅度逐渐减小;随雨强的增加同一边坡相同位置处土体含水率越早开始增大,其增长速率及幅度也随之增加;利用所建立的含水率遗传算法神经网络预测模型所得入渗结果与试验观测结果接近,表明该神经网络方法能较好地描述路基边坡土体含水率的变化情况及雨水的入渗过程。A series of rainfall tests with different rainfall intensities and different slope angles were carried out in laboratory to study the variation law of moisture content and the infiltration process in subgrade slope. According to the test results, the subgrade slope with the same engineering properties of soil, compaction degree and drainage condition were found, the moisture content was jointly influenced by four factors, i.e. rainfall duration, position of soil mass, slope angle and rainfall intensity. Taking the four factors as input units and the volume moisture content as output unit, the moisture content forecast model with genetic algorithm based on neural network was established. After prediction test, the neural network was used to forecast the infiltration process in slope with angle of 40° at rainfall intensity of 2.7 mm/min. The results show that as the rainfall lasts, the soil moisture content increases and the infiltrated area in the slope is extended. With the increase of distance from soil mass to infiltrate surface, the change of moisture content lags, and the increasing rate and amplitude of moisture content decrease. At the same rainfall intensity, the change process of moisture content at the top of the slopes with different slope angles is similar, whereas the increasing rate and amplitude of that at the toe of the slopes decrease with the increase of slope angle. When the slope angle remains unchanged, the increasing rate and amplitude of soil moisture content at similar position increase with the increase of rainfall intensity, and begins to change earlier. The forecasted infiltration process in slope is consistent with the observation result by rainfall test, which indicates that the neural network can exactly forecast the variation of moisture content and infiltration process in subgrade slope.

关 键 词：路基边坡 土体体积含水率 遗传算法神经网络 预测模型 入渗过程 

分 类 号：TU444[建筑科学—岩土工程]