论文部分内容阅读
针对缺乏非平稳变形(阶段性、反复性及突变性)特征数据导致的滑坡预测与评估不完全符合工程实际的情况,通过分析变形时序的物理意义与类型,在定义了非平稳变形趋势变化外延模式的基础上,提出了综合考虑滑坡当前变形阶段、观测变形数据特征以及待预测时段外界诱发因素的时序外延模式辨识方法,并进一步在工程先验知识指导下建立了支持向量机(SVM)预测模型。通过2个工程实例对方法进行应用验证。结果表明,工程先验知识能够有效补充非平稳变形的观测数据信息,对预测建模具有指导性作用;融入外延模式的SVM模型与一般性SVM的外推预测,其平均相对误差可降低2~3倍,预测的可靠性与准确性得到了显著增强。
The prediction and evaluation of landslides caused by the lack of non-stationary deformation (staged, repetitive and catastrophic) data do not fully meet the requirements of engineering practice. By analyzing the physical meaning and types of deformation sequences, Based on the model, a new method of time-series epitaxial pattern recognition considering the current deformation phase of the landslide, the characteristics of the observed deformation data and the external factors to be predicted is put forward and the support vector machine (SVM) prediction is further established under the guidance of engineering prior knowledge model. The method is verified by two project examples. The results show that the prior knowledge of engineering can effectively supplement the observed data of non-stationary deformation and play an instructive role in predictive modeling. The average relative error between SVM model and general SVM can be reduced by 2 ~ 3 times, the prediction reliability and accuracy have been significantly enhanced.