Please use this identifier to cite or link to this item:
Title: A Neural Network Based Rainfall Forecasting System Using Multiple Stations
Authors: Weerasinghe, H.D.P
Premaratne, H.L.
Sonnadara, D.U.J.
Keywords: Neural networks
Issue Date: 2007
Citation: Proceedings of the 25th National IT Conference (2007)
Abstract: This paper presents an implementation of a robust neural network based rainfall forecasting system based on a cluster of weather stations in the dry zone of Sri Lanka. The implemented network was based on a Feed-forward back propagation technique. A total of ten neighbouring weather stations having long periods of precipitation data were used to train and test the model. Twenty years of daily rainfall data were used to train the network and ten years of daily rainfall data were used to test the accuracy of the model. The model was trained separately to predict the rainy days during the North-East monsoon season. Instead of extracting two states such as ‘rain’ or ‘no rain’, the model was further improved to predict at many sub levels using Fuzzy Logic. The model consists of ten separate neural networks which were combined to form one model to predict the status of the daily rainfall at each station. When rainfall of previous three days in the nearest three neighbouring stations was taken into account, the rainfall occurrence model was able to make predictions within 72% - 83% accuracy. Predicting the North-East monsoon rainfall accuracy was within 68% - 76% and the fuzzy classification overall accuracy was within 75% - 87%.
Appears in Collections:Department of Physics

Files in This Item:
File Description SizeFormat 
2007 CSSL 04.pdf13.56 kBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.