Genetic Algorithm-Based Fuzzy Cluster Analysis for Flood Hydrographs

To determine typical flood hydrographs by making full use of observed data, the fuzzy c-means (FCM) clustering algorithm is used to cluster all observed hydrographs into several clusters, each of which gives a typical hydrograph. Because it is prone to fall into local optimum, especially when dealing with massive high dimensional data as in the flood hydrograph classification, the conventional FCM clustering algorithm is not fit for the hydrograph classification and is improved by means of genetic algorithms in this paper. The proposed approach is applied to classify the observed hydrographs in 40 years from 1959 to 1998 at Ningdu hydrological station, Jiangxi Province, China, and the clustering solutions are evaluated by using a cluster validity function based on possibility distribution theory. The results show that the split of two clusters of hydrographs at Ningdu station is the optimal solution, and the splits of 5, 6 and 7 clusters are also acceptable. The final clustering solutions give typical, realistic hydrographs in good accordance with the local hydrological and meteorological conditions, and that shows the proposed approach performs well.