類神經網路應用於道路邊坡落石坍方預測之可行性研究

A Feasibility Study of Applying the Artificial Neural Network to Prediction of Roadside Rockfalls and Landslides

鄭大偉(Ta-Wui Cheng) ； 王淑慧(Sue-Hwei Wang) ； 魏鎮東(Cheng-Tong Wei)

《鑛冶：中國鑛冶工程學會會刊》 45卷3期 (2001/09) Pp. 104-111+103

https://doi.org/10.30069/MM.200109_45(3).0011

摘要

道路邊坡發生落石坍方受到許多參數的影響，包括坡度、坡高、地質材料、地質構造、土地利用、雨量、地震等因素。一般道路邊坡的傳統評估方法只根據其簡單之線性關係來進行對落石坍方之評估。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，本研究利用類神經網路所具有之非線性及平行處理能力，來處理各項參數對落石坍方之影響，進行預測落石坍方的可能性。以台18線公路（阿里山公路）20公里至75公里為研究對象，其中隨機選取150筆案例為訓練資料，115筆為預測目標，預測準確率大約在80%左右，並比較統計分析中之判別分析預測落石坍方之能力，結果證明類神經網路具有預測落石坍方之能力且準確度高於判別分析之結果。

關鍵字

類神經網路 ； 落石 ； 坍方

並列摘要

The hazardous rockfalls and landslides by roadside are affected by many factors including the slope, slope height, geological material, geological structure, land use, rain and the earthquake, etc. Conventionally, the hazard of rockfalls and landslides by roadside is estimated according to the linearly relation. Furthermore, the influences of each factor are not linearly related. In order to solve the above problems, this study takes advantage of the non-linear and parallel attributes of the artificial neural network to process each factors, which are coritributed to rockfalls and landslides, and subsequently to predict the possibility of their occurrence. The section from 20 km to 75 km on Tai.18 Highway (A.Ii.shan Highway) was chosen for the research object. 150 cases were randomly selected as the training data, while other 115 cases as the prediction targets. The results prove that artificial neural network does have the ability to predict rockfalls and landslides, and the accuracy of the prediction can be reached 80%. Finally, this study also emphasizes the effects introduced by different amount of rainfall in order to discover the relationship between rainfall and the road section suffered from rockfalls and landslides. This can facilitate the rescue operation in the event of a road disaster, and also provide as a reference for the road construction.

並列關鍵字

Artifical neural network ； Rockfall ； Landslide