基于地表监测数据和非线性时间序列组合模型的滑坡位移预测

Landslide displacement prediction based on surface monitoring data and nonlinear time series combination model

针对滑坡位移–时间曲线的非线性特征和以往预测模型的不足,提出基于地表监测数据和非线性时间序列分析的组合模型预测滑坡位移。以新滩滑坡和三舟溪滑坡为例,通过对位移、降雨、库水位等资料的分析,研究滑坡的变形特征及影响因素。在利用逆序法和小波分析检验滑坡位移的趋势特征和周期特征的基础上,采用非线性组合模型进行预测,包括利用多项式拟合并预测趋势项位移;用一种基于小波分析的三角函数法(WA-TF)对周期项位移进行预测;遗传算法优化选参的BP神经网络(GA-BP)对波动项位移进行预测。最终将各位移分量累加得到滑坡的累积位移预测值,并与监测值进行对比分析。结果表明非线性组合模型的预测精度高且具有较好的通用性,为滑坡位移定量预测提供了一种可行的思路。

According to the nonlinear characteristic of displacement-time curve of landslide and deficiency of traditional prediction models,the combination model based on surface monitoring data and nonlinear time series analysis was proposed to predict landslide displacement. Taking Xintan landslide and Sanzhouxi landslide for example, deformation characteristics and influence factors were researched by analyzing information of displacement,rainfall and reservoir water level and so on. On the basis of testing trend and periodic characteristics of landslide displacement by using athwart order method and wavelet analysis,nonlinear combination model was used to predict it. Trend item displacement was fitted and predicted by polynomial function. Using a trigonometric function based on wavelet analysis（WA-TF model） to predict periodic item displacement. BP neural network whose parameters were optimized by genetic algorithm was used to predict random item displacement. The accumulative displacement was obtained by adding up all items displacement and it was compared and analyzed with the monitoring values. The results indicate that nonlinear combination model has high accuracy and good versatility. This provides a possible thinking for quantitative prediction of landslide displacement.