運用空間知識探勘技術在土石流分類模式建立之研究－以陳有蘭溪流域為例

The Comparison Study on Debris-Flow Hazard Knowledge Inference Models Based on Spatial Data Mining Techniques-A Case Study of Chen-Yu-Lan Stream

雷祖強(Tsu-Chiang Lei) ； 萬絢(Shiuan Wan) ； 林沂樺(Yi-Hua Lin) ； 吳仕傑(Shih-Chieh Wu) ； 李素馨(Su-Hsin Lee)

《水保技術》 7卷4期 (2012/10) Pp. 198-211

https://doi.org/10.30040/JSWCT.201210.0001

摘要

本研究欲使用空間知識探勘技術建立一套新的土石流潛勢溪流分析模式，研究流程如下：1. 蒐集陳有蘭溪流域內地文因子以及水文因子等二十一項因子，以克利金法（Kriging）與GIS技術建立陳有蘭溪土石流監測資料庫；2.在描述土石流發生因子中，透過主成份分析（Principal Component Analysis, PCA）和粗糙集合理論（Rough Set, RS），針對土石流因子進行萃取，並對土石流發生描述之有效性進行探討，以避免不必要的雜訊被引入至資料庫當中；3.運用線性多變量分析（Linear Discriminant Analysis, LDA）與非線性之支援向量機（Support Vector Machine, SVM）兩種模型，進而對資料進行分類與辨識。研究成果顯示，RS+SVM模型在土石流災害分類上的能力最好，而PCA+LDA模型的能力較差，這也顯示了目前土石流發生推論機制仍充滿了高度的不確定性與預測的困難性。

關鍵字

土石流 ； 知識推論 ； 線性分類 ； 資料挖掘 ； 非線性分類

並列摘要

In this study, we used different ”Data Mining” process to determine the available classification models for characterization of debris-flow problems through the spatial knowledge techniques. Therefore, in this study have three steps: 1. To collect the complete data of study site into the debris-flow data warehousing (21 factors). Many kinds of different scaling data through Kriging and other GIS techniques are applied; 2. Two approaches of Principal Component Analysis (PCA) and Rough set (RS) methods can improve the classification accuracy on the debris-flow issues through extracted core factors of the variables. 3. The development of debris-flow knowledge inference engine are two models of (1) Statistical of Linear Discriminate Analysis+ Principle component analysis (LDA) and (2) Nonlinear classifier of Support Vector Machines (SVM). Finally, the results of study, the RS+SVM model have the best debris flow knowledge inference ability of classification results, and linear multivariate analysis of the PCA + LDA model has the inadequate ability of classification of debris flow problem.

並列關鍵字

Debris-Flow ； Knowledge Inference ； linear classifier ； data mining ； nonlinear classifier