落石邊坡危險度與危害度分級與預報

Abstract

傳統落石坡危險度評估模式中，權數的給定都是來自專家經驗累積所得，這樣的結果也造成權數的給定會因人而異，且主觀意識強烈。透過類神經網路學習所得的參數較具公信力，網路架構因學習案例的累積更形建全，使用上不會像傳統評估方法因人的主觀因素而影響其功能。本研究以中橫谷關-德基段、新中橫同富-塔塔加段為研究範圍，台八線中橫谷關-德基段因921集集大地震現已封閉暫停使用，台八甲線也嚴重受損，台21線同富-塔塔加段則是在121.2k、136.8k、146.2與147.7k附近嚴重毀壞。 本文針對這兩個路段加以危險度與危害度等級評估，透過類神經網路系統在案例學習與預測都有顯著的成果。另外亦以地質因子加上兩項雨量因子構成另一因降雨引致落石坡危險度評估模式，透過這套模式與雨量災害案例的學習，將來配合氣象局發布豪大雨預報進行管理路段的危險度評估，必要時，道路管理者可因危險路段過多應該將道路暫時封閉或示警，以求將災害降至最低。

Based on the experience, the professional engineer provides the rating weights of the influenced parameters in the rockfall hazard rating systems. Therefore, the given rating weights will strongly depend on anybody’s subjective sense. One can obtain more conviction in the results of learning by using Artificial Neural Network (ANN). The network framework will be perfect by increasing of the learned cases. ANN is not alike traditional hazard rating system that will affect the function in using. The objects of this study are the areas along the Central Cross-Range Highway and the New Central Cross-Range Highway. The former area had been destroyed seriously and was temporarily closed due to 921 Chi-Chi Earthquake in 1999. The later area had been destroyed nearby 121.2k、136.8k、146.2k、147.7k. It presented the outstanding results of learning and rating for two areas through ANN. Furthermore, the another ANN model includes the geologic factors and two rainfall parameters will predict the hazard of raining for all slopes. Based on the ANN model, the rockfall hazard ranks can be determined following the weather forecast from the Central Weather Bureau. Also, the road manager should temporarily stop the road users to get into the area if it had been destroyed seriously.