时间序列分析与支持向量机的滑坡位移预测

Time series analysis and support vector machine for landslide displacement prediction

滑坡在变形演化过程中,遭受季节性外界影响因素的作用,变形位移时间曲线呈现出阶跃型特征.采用时间序列分析方法,将位移分解为趋势项和季节项.趋势项位移由坡体自身地质条件控制,利用多项式函数进行预测;季节项位移受降雨、库水位和地下水位等因素的季节性作用而变化.选取当月降雨量、累计前2个月降雨量、当月库水位高程、月库水位变化速率和当月地下水位高程作为影响因子,利用进化支持向量机耦合模型进行预测;通过时间序列加法模型得到滑坡总位移预测值.以三峡库区白家包滑坡为例,通过计算得到预测结果与实际监测值基本吻合,其中最大均方根误差为18.8,而最小相关系数为0.98.研究表明:基于时间序列分析与进化支持向量机的滑坡位移预测模型,有效反映了阶跃型滑坡位移变化规律与季节性影响因素之间的响应关系,是一种行之有效的滑坡位移预测方法.

For the deformation and evolution process of landslides, the displacement curve presents the step-type deformation pattern under the influence of seasonal external factors＇ effect. The time series analysis method was used to separate the landslide displacement into trend term and seasonal term. Trend term displacement was controlled by the geological co^aditions of the slope and to be predicted using displacement polynomial function. Seasonal term displacement was affected by the seasonal functioning of influential fac- tors, such as rainfall, reservoir level fluctuation, underground water level and so on. The rainfall of cur- rent month, cumulative rainfall of anterior two months, reservoir level fluctuation of the current month, the month rate of change of water level and underground water level of the current month were selected as influential factors. The genetic algorithm and support vector regression were adopted to predict the season- al term displacement. Then, the prediction total displacement was obtained by the time series additive model. The Baijiabao landslide in the Three Gorges reservoir area was selected for a case study. The predicted values of the model were consistent with the measured values. The maximum value of mean squared error was only 18.8, and the minimum value of correlation coefficient up to 0.98. The results show that the proposed method effectively reflects the relationship between landslides deformation and the external influential factors. It is an effective method for prediction of the step type deformation landslides.