基于前期累积降雨和高斯过程回归模型的滑坡位移预测

Prediction of landslide displacement based on previous accumulated rainfall and Gaussian process regression model

降雨是诱发滑坡变形的重要因素,传统的位移时间序列模型预测滑坡位移时,仅考虑历史位移而未考虑降雨影响,致使中长期或长期位移预测误差较大。鉴于高斯过程回归(GPR)算法具有易于实现、超参数自适应获取及预测输出具有概率意义等优点,通过引入累积降雨量指标,建立了考虑前期累积降雨的滑坡位移GPR预测模型,提升了模型长期位移预测的能力。以西昌南采场滑坡为例,首先对滑坡位移-日累积降雨量曲线进行分析,并采用皮尔逊法计算了滑坡位移与前期累积降雨天数之间的相关性系数;其次,建立GPR模型,利用已有监测数据对模型进行训练和测试验证,结果表明所建模型相比不考虑前期累积降雨的模型预测精度大大提高。在此基础上,对监测点S1-1和S1-2开展长期位移趋势预测,并对比分析了在假设增加10%和20%降雨的情况下的位移变化趋势,结果表明增加20%降雨导致滑坡变形速率增加至17 mm/d左右,未采取治理措施情况下滑坡会出现加速滑移失稳。

Rainfall is an important factor inducing landslide deformation.Traditional displacement time series prediction models only consider historical displacement and do not consider the impact of rainfall to conduct landslide displacement prediction,resulting in significant errors in the medium to long term or long-term displacement prediction.Given that the Gaussian process regression(GPR)algorithm has the advantages of easy realization,adaptive acquisition of hyperparameter and probability significance of prediction output,a GPR prediction model of landslide displacement considering the previous accumulated rainfall by introducing the cumulative rainfall index is established in this study,which improves the long-term displacement prediction capability of the model.Taking the landslide in the southern mining area of Xichang as an example,the relationship curve of landslide displacement and daily cumulative rainfall was first analyzed,and the Pearson method was used to calculate the correlation coefficient between the landslide displacement and the previous cumulative rainfall days;Secondly,a GPR model was established,and trained and tested using existing monitoring data.The results showed that the prediction accuracy of the established model was significantly improved compared to the model without considering the previous accumulated rainfall.On this basis,a long-term displacement trend prediction was conducted for monitoring points S1-1 and S1-2,and a comparative analysis was performed on the displacement trend under the assumption of increasing rainfall by 10%and 20%.The results indicate that a 20%increase in rainfall leads to an increase in the deformation rate of the landslide to about 17 mm/d.Without taking control measures,the landslide will experience accelerated sliding to instability.