Author(s): C. L. Wu; K. W. Chau
Linked Author(s):
Keywords: Daily rainfall prediction; Singular spectral analysis; Fuzzy c-means clustering; Hybrid ANN-SVRmodel
Abstract: A hybrid model integrating artificial neural network (ANN) and support vector regression (SVR) was developed for daily rainfall prediction. In the modeling process, singular spectrum analysis (SSA) was first adopted to filter the raw rainfall data. Fuzzy c-means (FCM) clustering was then used to split the training set into three crisp subsets (low, medium and high levels) according the magnitude of the rainfall data. Two local ANN models were involved in training and predicting the low-and medium-level subsets whereas a local SVR model was applied to the high-level subset (the hybrid model is referred to as ANNSVR-SSA). A conventional ANN model was selected as the benchmark. The ANN with the SSA was also studied (referred to as ANN-SSA). These models were examined by two daily rainfall series from China at1-day-, 2-day-, and 3-day-ahead forecasting horizons. The comparison results showed that the ANN-SVRSSA, among the three models, had the best performance. In addition, the ANN-SSA exhibited considerable accuracy in rainfall forecasting compared with the ANN.
Year: 2009