Improving prediction of dam failure peak outflow using neuroevolution combined with K-means clustering
Journal of Hydrologic Engineering
American Society of Civil Engineers
Estimation of peak outflow resulting from dam failure is of paramount importance for flood risk analysis. This paper presents a new hybrid clustering model based on Artificial Neural Networks and Genetic Algorithm (ANN-GA) for improving predictions of peak outflow from breached embankment dams. The input layer of the ANN-based model comprises height and volume of water behind the breach at failure time plus a new parameter of ‘cluster number’. The cluster number is obtained from partitioning the input data set using K-means clustering technique. The model is demonstrated using the data samples collected from the literature and compared with three benchmark models by using cross-validation method. The benchmark models consist of a conventional regression model and two ANN models trained by non-linear techniques. Results indicate that the suggested model is able to estimate the peak outflows more accurately especially for big flood events. The best prediction for the current database was obtained from a five-clustered ANN-GA model. The uncertainty analysis shows the five-clustered ANN-GA model has the lowest prediction error and the smallest uncertainty.
The authors gratefully acknowledge the financial and other supports of this research provided by the Islamic Azad University, Islamshahr branch, Tehran, Iran
This is the author accepted manuscript. The final version is available from the publisher via the DOI in this record.
DOI: 10.1061/(ASCE)HE.1943- 5584.0001505