基于多元自回归样条的液化侧移灾害评估模型

An Estimation Model of Liquefaction-induced Lateral Spread Based on a Multivariate Adaptive Regression Spline

摘要:

为了将液化侧移计算方法本土化，本文基于Youd等人收集的液化侧移数据库，采用擅长处理复杂参数维度问题的多元自适应样条回归法（MARS）建立液化侧移计算模型。经验证，该方法建立的模型精度和可靠性都可以通过检验。本文主要内容包括：（1）在本土化过程中，本文选取峰值加速度PGA作为地震动参数，替代现有模型中的M和R，计算结果并未影响液化侧移计算精度；（2）在MARS方法进行建模过程中，对临空和缓坡两种情况的影响因素进行重要性程度计算和筛选；（3）建立两种情况下的液化侧移计算模型，并用F检验法和t检验以及新西兰2010—2011年系列地震中液化侧移数据进行检验，证明本文方法的合理性和可靠性。本文的结果能满足我国相关规范、工程师以及岩土地震工程防灾减灾等方面的需求，还可为地震液化侧移灾害小区划提供理论基础。

Abstract:

To build a liquefaction-induced lateral spread estimation model, the multivariate adaptive regression spline (MARS) method, which can better deal with multidimensional parameters, is used in this paper. Through verification, we found the accuracy and reliability of the model bears examination. In this paper, (1) in the process of localization, PGA was selected as the seismic parameter to replace M and R in present models, and the calculation accuracy was not affected; (2) in modeling with MARS method, the impacts of various parameters were studied, and the main influencing factors under the two conditions of free face and mild slope were screened; (3) two estimation models were built under the two conditions, then tested by the F and t testing method using the liquefaction-induced lateral spread data from Christchurch earthquakes during 2010-2011. The rationality and reliability of the method were proved. Results of this paper can satisfy the demands of disaster prevention and reduction engineering of geotechnical earthquake engineering, and can also provide a theoretical basis for the disaster zoning of liquefaction-induced lateral spread.