摘要
为提高基于时间序列分析的滑坡变形预测模型的精度及实用性,提出一种基于灰色最小二乘支持向量机的滑坡变形预测算法。算法应用灰色理论对滑坡变形原始序列累加以增强数据规律性,利用最小二乘支持向量机良好的非线性拟合特性,建立灰色最小二乘支持向量机(GLSSVM)模型,并采用遗传算法选取最优参数。以三门洞滑坡为例,对该滑坡变形监测值与预测值进行对比分析。结果表明:此模型在基于时间序列的滑坡变形预测上能保证较高的精度,验证了该模型在滑坡变形预测上应用的可靠性。依据地质特征和计算结果对滑坡进行预测,预测认为2019年1月至2021年12月三门洞滑坡整体表现为牵引式变形特征,形成整体滑移的可能性较大;但变形位移方向局部存在差异,滑坡东南部中前缘高地变形滑移方向与滑坡主滑方向(东偏北28.5°)明显不一致,向西北方向(61.5°)滑移。
In order to improve the accuracy and practicability of landslide deformation prediction model based on time series analysis,a landslide deformation prediction algorithm based on Grey Least Squares Support Vector Machine(GLSSVM)was proposed.Firstly the Grey Theory was applied to enhance the regularity of the original sequence of landslide deformation data by summation,then using the good non-linear fitting characteristics of least squares support vector machine,a GLSSVM model was established,and the genetic algorithm was used to select the optimal parameters.Taking Sanmendong Landslide as an example,the monitoring and predicting values of landslide deformation were compared and analyzed.The results showed that the model can guarantee high accuracy in landslide deformation prediction based on time series,and verify the reliability of the application of the model in landslide deformation prediction.According to the geological characteristics and calculation results,it was predicted that from January 2019 to December 2021,the whole Sanmendong Landslide was characterized as traction deformation,and the possibility of overall slip in the future was greater,but there were local differences in the direction of displacement of different parts.The sliding direction of middle front high ground in the southeast of the landslide were different from the main slip direction(28.5°NE),whose displacement slips northwest(61.5°NW).
作者
袁乾博
肖诗荣
李春霞
夏宇
王涛
YUAN Qianbo;XIAO Shirong;LI Chunxia;XIA Yu;WANG Tao(College of Civil Engineering&Architecture,China Three Gorges University,Yichang 443002,China)
出处
《人民长江》
北大核心
2020年第7期130-135,共6页
Yangtze River
基金
国家自然科学基金项目“靠椅状顺层岩质水库滑坡机理及空间预测模型研究”(41272310)。
