離散粗集合在土石流潛勢溪流特性之研究－以陳有蘭溪流域為例

The Application of Discrete Rough Set on Potential Features of Debris Flows: A Case Study on Chen-Yu-Lan River

萬絢(Shiuan Wan) ； 雷祖強(Tsu-Chiang Lei) ； 陳政宇(Z.Y. Chen)

《水保技術》 3卷2期 (2008/03) Pp. 73-86

https://doi.org/10.30040/JSWCT.200803.0003

摘要

土石流對台灣災害非常嚴重，然而，土石流的水文和地文因子收集有賴於地理資訊系統(GIS)、數值高程模型(DEM)與衛星感探測(SPOT)等相關技術取得。本研究使用上述的相關空間技術並整合布林粗集合理論來達到資訊化約的目的，透過此種新的化約方式能夠在模糊與不確定的土石流中有效的擷取資訊，並透過決策屬性的建立有效將影響因子萃取出來，以建構影響土石流之危險因子。此種方式極為合適用於混淆問題中分類之所需。
本研究以交叉驗證法，從陳有蘭溪土石流的資料庫219筆中，反覆隨機選取20筆訓練資料，利用粗集合理論建立知識規則，並以剩餘的199筆測試範例判釋土石流的正確率，直到得到20筆最佳訓練資料，即最高的判釋土石流的正確率，依其將真值量化後之離散化算法，求出重要核心因子，獲得知識界限圖（門檻值），並利用其知識界限對其餘199筆進行測試。本研究的貢獻在(1)從陳有蘭溪流域中所建構之龐大的資訊系統中，找出土石流判釋率之規則。(2)經由所建立之知識規則庫，可得知最重要的影響因子依次為集水區面積、主流平均坡度、河川頻率等重要屬性因子。(3)建立土石流災害潛勢圖，對於土石流判釋有一定分類之成效，以便能快速、即時提供給專家學者們對土石流災害監測、參考之用。本研究判釋成果顯示，以布林粗集合所擷取到的重要因子來提昇土石流判釋準確度，布林粗集合顯示的分類準確度達72%。布林粗集合的優點在於以很少的資訊樣本即可建立一定準確度的土石流判釋的知識規則。

關鍵字

布林粗集合理論 ； 數值高程模型 ； 集水區面積 ； 主流平均坡度

並列摘要

Debris flow is very serious to the calamity of Taiwan. However, the geomorphology and hydraulic factors of debris flows are collected through relevant technology such as GIS, DEM and SPOT ...etc. This research uses above-mentioned spatial information technology and Boolean Rough Set to extract the related spatial information. Most of the current research relied on this new technology to extract the information of chaotic data in debris flow. The relations between attributes and decision were established through Boolean Rough Set. Then, the influence factors and knowledge rules for debris flow are found.
The study also used Cross-Validation to randomly choose the 20 numbers training data set from the database of debris flow hazard area of Chen Yu Lan River. These training dataset will attain the knowledge rules of the debris flow of occurrence/non-occurrence. The process of selecting training dataset will not stop until the maximum accuracy rate of the testing dataset approach. The Boolean Rough Set algorithm was developed to computer program to find threshold and core attributes. The knowledge scope diagram is generated by using threshold values. The contributions of this study are shown as: (1) The research can find the optimum accuracy rate of debris flow from the Chen-Yu-Lan database. (2) Watershed area, Average of primary slope, Stream frequency are the major influenced factors for debris flows. (3) Aftermath, the debris flows potential diagrams are drawn and a rational accuracy rate of landslide is calculated. Thus, an instant and on-time monitoring and management of debris flows strategy can be made. The contribution of this study shows that Boolean Rough Set could attain the accuracy rate for debris flow of 72%. The advantage of using Boolean Rough Set can render the knowledge rules of the occurrence of the debris flow.

並列關鍵字

Boolean Rough Set ； DEM ； Watershed area ； Average of primary slope