應用Scoops3D進行滑動面位置之最佳化與崩塌機率之分析

Abstract

坡地災害是台灣常見的天然災害之一，致災因子、災害發生的機制與災後整治工程(或措施)的效益評估，因此成為本研究關著的重點。鑒於坡地災害本質上是以三維的運動方式，本研究首先利用三維模式Scoops3D進行分析、之後輔以二維模式Geostudio探討整治工程(或措施)的效應，其中，諸多因子的不確定性均納入分析的過程中，如雨量、雨型、地下水位、土壤參數與沖刷深度等。 在崩塌發生之前，通常會產生地表特徵如坡頂裂縫或邊坡位移，本研究參考這些地表與地中特徵資訊，應用Scoops3D和粒子群演算法(PSO, particle swarm optimize )推算最佳化滑動面位置，由於選用不同的BPNN訓練架構和最佳化之目標函數，最後挑選出整體誤差較小的11個滑動面，其中3個滑動面的幾何位置近乎一致，且崩塌機率較高，因其最接近真實滑動面位置，故推測為最佳滑動面。 三維模式Scoops3D無法考慮整治措施對邊坡的影響，故本研究應用二維模式Geostudio 建立5個剖面的二維模型，分析各剖面邊坡在延時48小時、重現期50年降雨下的崩塌機率，再以面積加權的方式計算最佳化滑動塊體的崩塌機率，計算結果為0.1956，而應用三維模式Scoops3D分析最佳化滑動塊體的崩塌機率為0.113，兩者的分析結果相近，且如預期地，二維模式的分析結果偏保守側，故初步推斷以二維模式推估整治後三維滑動塊體崩塌機率為可行的方法。 另外再應用Geostudio分析兩個剖面(SB-6剖面和18-5W剖面)邊坡在沖刷深度和重現期降雨下對崩塌機率的影響，分析結果為兩個剖面的崩塌機率皆會受沖刷深度和重現期降雨的影響，其中沖刷深度的影響又更為顯著。

Slope disasters are one of the common natural disasters in Taiwan. The hazard factors, disaster occurrence mechanism, and effects of post-disaster remediation projects have therefore become the focus of this research. In view of the fact that slope disasters are essentially three-dimensional movements, this study first used the three-dimensional model, Scoops3D, for analysis, and then supplemented with the two-dimensional model Geostudio to discuss the effects of remediation projects. Among them, the uncertainties of many factors were included in the analysis, such as rainfall, rain type, groundwater level, soil parameters, scouring depth and so on. Before landslide, observations such as tension cracks or displacement of the slope are usually found, so this study first utilizes Scoops3D and particle swarm optimization (PSO) algorithm to optimize the position of the slip surface with the collected information. Due to different BPNN training structures and objective functions of optimization, 11 candidates of slip surface are selected, in which 3 of 11 slip surfaces are almost at the same position, indicating this position possesses a higher landslide occurrence probability, so it is selected and used for further analyses. The three-dimensional model Scoops3D cannot consider the retrofitting method, the two-dimensional model Geostudio is therefore, used to analyze landslide probability with the retrofitting. Several cases are analyzed, for example, to compare the difference between Scoops3D and Geostudio, a case of 50-year returning period and 48-hour rainfall event is considered. Results shown that the landslide probabilities are 0.1956 and 0.113 for Geostudio and Scoops3D, respectively. As expected, the 2D model delivers a similar but more conservative outcome, indicating the proposed 2D analysis framework can provide a reasonable estimation of landslide occurrence probability. In addition, Rainfall has been long recognized as one of the most significant triggering factors for slope failures, Geostudio was used to analyze (slope of SB-6 profile and slope of 18-5W profile) the landslide probability under different scour depths and return-period rainfall events to further illustrate their influence on landslide occurrence probability.