基于遗传算法-支持向量机的滑坡渗透系数反演

求解库岸边坡岩土体的渗透系数是研究滑坡渗流场及多场演化的基础，一般通过原位试验和室内试验求得，但试验成本较高且试验位置具有一定的随机性。本文以三峡库区马家沟滑坡为例，提出一种利用地下水位动态观测资料反演滑坡岩土层渗透系数的方法。具体步骤为:（1）依据滑坡的勘察资料和水位观测数据，构建滑坡数值模型；（2）利用SPSS生成不同渗透系数正交试验组合，并将渗透系数代入数值模型中计算监测井的水位，得到不同渗透系数及其对应的模拟水位数据；（3）应用遗传算法优化的支持向量机构建坡体模拟水位与渗透系数的非线性映射关系，再通过代入实际动态监测水位值求得滑坡岩土层的渗透系数；（4）将求得的渗透系数代入数值模型，用计算的模拟水位与实际观测水位进行对比验证。研究结果表明:遗传算法优化的支持向量机具有良好的学习预测效果，能准确预测渗透系数与水位的关系。该反演方法具有高效、准确的优点，反演结果的精度满足实际应用需要。

Estimation of hydraulic conductivity of landslides based on support vector machine method optimized with genetic algorithm

Estimation of hydraulic conductivities (K) of the rock media in a landslide is the basis for the study of the seepage field and multi-dimensional evolution of the reservoir bank slope. Traditionally, in-situ tests and indoor tests are used to determine the hydraulic conductivity of landslide rock and soil, but this method is costly and the test location has a certain randomness. In this study, the Majiagou landslide in the Three Gorges Reservoir area is taken as an example, and a method for inverting the K values of the deformed rock and soil mass using the groundwater level dynamic monitoring data is proposed. The basic idea is as follows. First, build a numerical model of the landslide based on the landslide survey data and water level observation data. Afterwards, SPSS is used to generate different orthogonal test combinations of hydraulic conductivity, substitute the hydraulic conductivity into the numerical model to calculate the water levels of the monitoring wells, and obtain the data of hydraulic conductivity and corresponding simulated water levels. Finally, the support vector machine (SVM) optimized with the genetic algorithm (GA) is used to construct a nonlinear mapping relationship between slope water level and hydraulic conductivities (K). The results obtained are then replaced for the monitored water levels to obtain the hydraulic conductivities of the landslide rock and soil which is used to develop the finite element model. The model is then verified by comparing the simulated water levels with the observed water levels. The inversion of the Majiagou landslide hydraulic conductivity shows that the SVM optimized with GA yields a good agreement between the simulated and real data and has a very efficient and accurate search results. The inversion accuracy of K based on the GA-SVM method meets the needs of practical applications.