Back to Search
Start Over
Discretized and Continuous Target Fields for the Reservoir Release Rules During Floods.
- Source :
- Water Resources Management; Sep2012, Vol. 26 Issue 12, p3457-3477, 21p
- Publication Year :
- 2012
-
Abstract
- This paper employed two classical, popular decision-tree algorithms (C5.0 and CART), and traditional Regression to deal with reservoir operations regarding decision of the releases from a reservoir system during floods. The experiment site was in Shihmen Reservoir, located in northern Taiwan. In a typical single-peak typhoon, the rules derived include two operational stages, the stage before peakflow (Stage I) and the stage after peakflow (Stage II). This study collected 50 typhoons (1987-2009). Four cases are designed, that are discretized class labels (target fields) are run by C5.0 and CART (i.e., Cases 1 and 2, respectively), while numeric class labels are run by CART and Regression (i.e., Cases 3 and 4, respectively). The criteria of root mean square error (RMSE), coefficient of efficiency (CE), and relative error of peak discharge (EQ) were used to evaluate the forecasts. Results showed that the decision trees are skillful in the prediction of reservoir releases in the studied site. Furthermore, it was found that CART regression trees with numeric targets are more appropriate and precise than C5.0 classification trees and Regression for the prediction of releases. [ABSTRACT FROM AUTHOR]
- Subjects :
- FLOODS
RESERVOIRS
ALGORITHMS
DECISION trees
Subjects
Details
- Language :
- English
- ISSN :
- 09204741
- Volume :
- 26
- Issue :
- 12
- Database :
- Complementary Index
- Journal :
- Water Resources Management
- Publication Type :
- Academic Journal
- Accession number :
- 78322252
- Full Text :
- https://doi.org/10.1007/s11269-012-0085-2