纵剖面形态的黄土斜坡地震危险性Logistic回归模型∗

Logistic Regression Model for Seismic Risk of Loess Slopes Considering Longitudinal Profile Morphology

摘要:

强震诱发斜坡失稳破坏是黄土地区主要震害之一，快速准确判别黄土斜坡地震危险性对城市抗震规划和震后应急救援具有重要意义。利用野外实地调查的 1920 年海原地震诱发 620 组滑坡和 380 组未滑斜坡数据，验证纵剖面形态对黄土斜坡地震危险性的影响，对斜坡纵剖面形态进行单因素 wald 检验，单因素 wald 检验结果表明纵剖面形态对于滑坡地震危险性快速判别有积极意义。利用野外调查中简单易得的坡高、坡角、纵剖面形态和地震烈度几个因子，建立黄土高原区斜坡地震危险性判别的 Logistic 回归模型，给出黄土斜坡地震危险性的快速判别公式，使用该公式对 1920 年海原地震诱发滑坡回判和 1718 年通渭地震诱发黄土滑坡校验，同时对比了不考虑纵剖面形态的 logistic 回归模型的结果。计算表明：利用 Logistic 回归模型快速判别黄土斜坡地震危险性，合理确定影响因子并调整 Logistic 回归中各影响因子的参数，计算得到的斜坡失稳与实际滑坡发育情况较为吻合；斜坡纵剖面形态是影响斜坡地震危险性的重要影响因素，在 Logistics 回归模型中考虑该因素得到的预测公式，判别斜坡失稳准确度比不考虑该因素时高 3% 左右。

Abstract:

The instability and destruction of slope caused by seismic activity is one of the main earth? quake damages in the loess area. It is of great significance for urban seismic planning and post-earth? quake rescue to determine the seismic hazard of slopes quickly and accurately. The field survey data including 620 groups of landslides and 380 groups of stable slopes during the 1920 Haiyuan earth? quake were used to verify the influence of the longitudinal profile shape on the seismic hazard of loess slopes using the single-factor wald test. The single-factor wald test results indicated that the longitudi? nal profile of the slope had positive significance for the rapid identification of the seismic hazard of landslides. Using the easy-to-obtain factors such as slope height， slope angle， longitudinal profile shape and seismic intensity in the field survey， a Logistic regression model for determining the seis? mic hazard of slopes in the Loess Plateau region was established， and a fast determining formula for the seismic hazard of loess slopes was given. This formula was used to verify the regression of the landslide induced by the Haiyuan earthquake in 1920 and the loess landslide induced by the Tongwei earthquake in 1718. At the same time， the results of the logistic regression model without considering the shape of the longitudinal section were compared. The calculations show that：1. The seismic haz? ard of loess slopes can be quickly distinguished using the Logistic regression model by reasonably de? termining the influencing factors and adjusting the parameters of each influencing factor in the Logistic regression. The calculated slope instability characteristic is more consistent with the actual landslide evolution；2. The slope longitudinal profile shape is an important factor that affects the seismic risk of slopes. The prediction formula obtained by considering this factor in the Logistics regression model can produce about 3% more accuracy in identifying the slope instability than that without considering this factor.