基于进化支持向量机的滑坡地下水位动态预测

Prediction of ground water level in landslides based on genetic-support vector machine

地下水位动态预测对滑坡稳定性评价具有关键作用。滑坡地下水位演化过程是一个受水文地质条件控制,并受降雨、库水和气温等多种影响因素综合作用而发展演化的非线性动力系统,地下水位与其影响因素之间存在非线性响应关系。以三峡库区白家包滑坡地下水位监测数据为例,在深入分析滑坡地下水位变化特征及其与影响因素响应关系的基础上,利用非线性智能遗传算法和支持向量机建立进化支持向量机耦合模型,并对地下水位进行预测,其预测结果的均方差和相关系数的平方分别为0.013和0.929,说明预测结果与实测值较吻合。选择神经网络模型进行对比,耦合模型的均方差小154%,而相关系数的平方大10%。综合表明进化支持向量机耦合模型具有较好的拟合和泛化能力,是一种行之有效的滑坡地下水位预测方法。

Prediction of ground water level is ignificant in evaluation of landslide stability. The evolution process of the ground water level in landslides is a nonlinear dynamic system which is controlled by the hydrogeology condition and to suffers from comprehensive effects by multiple influential factors such as rainfall, reservoir water level, temperature and so on. There is the nonlinear response between ground water level and its influencing factors. According to ground water level data of Baijiabao landslide in the Three Gorges reservoir area, the response relationship between influential factors and ground water level variation was analysed, and the characteristics of ground water level in the landslide were discussed. Using a nonlinear genetic algorithm and support vector regression （GA-SVR） model, the values of ground water level in landslides is predicted. Predicted values of the GA-SVR model are consistent with the measured values. The mean squared error of the GA-SVR model is only 0.013, which is less than those of radial basis function artificial neural network （RBF-ANN） model by 154%. And the squared correlation coefficient of the GA-SVR model reaches 0.929, which is more than those of RBF-ANN model by 10%. It is indicated that the GA-SVR model has a great fitting and generalization ability. It is an effective method for prediction of ground water level in landslides.