類神經網路於落石坡危險度評估

Abstract

落石坡的危險度受到許多參數的影響，包括地形、地質等因素。一般落石坡的傳統加權評估法只根據其簡單之線性關係來進行對落石坡危險度之評估。但是，其每個參數對落石坡的影響並不是屬於線性關係，因此並非十分的適合。因此，本論文利用類神經網路所具有之平行處理能力，來處理各項參數對落石坡危險度之影響，進行對其危險度之分析。本研究以中橫谷關－德基水庫間的237 個落石坡為調查對象，其中隨機選取 150 個落石坡為訓練資料，87 個落石坡為預測目標，利用類神經網路程式加以分析預測，分析結果證明了類神經網路具有評估落石坡危險度之能力。並利用類神經網路程式對 237 個落石邊坡預測其於發生暴雨後之落石量及建立標準評估表，以為其他路段或調查之用。

Rock falling is induced by several factors. Factors include slope geometry, slope geology, and others. Based on the assumption of simple deterministic relation between the influenced factors and the potential of rockfall hazards, rock hazard rating systems or similar rating criteria and scoring methods were developed and popularly adopted for assessing the potential of rockfall hazards. However, the relations between the factors and rockfall hazards are complicated and highly nonlinear. The assessments of rock fall hazards for potential slopes using such systems may be subjective and unreliable. In this thesis, an artificial neural network (ANN) model is proposed for assessing the rockfall potential. 237 instances of potential rockfall sites along the Central Cross-Range Highway are collected to evaluate the feasibility of the new model, in which 150 instances are used to training the ANN and the other 87 instances are used to test the performance of the new model. The results are compared with that obtaining from the conventional methods. This thesis present the available assessment that results to demonstrate the efficacy of the new approach. This trained ANN can be applied to predict the potential of rockfall hazard iin other areas.