Water level predictio for Limbang basin using multilayer perceptron (mlp) and radial basis function (rbf) neural network

• Main Author: Muhammad Noor Hisyam, Abg Hashim
• Format: Final Year Project Report
• Language: English; English
• Published: Universiti Malaysia Sarawak (UNIMAS) 2010
• Subjects: QC Physics; TC Hydraulic engineering. Ocean engineering
• Online Access: http://ir.unimas.my/id/eprint/7814/1/Noor%20Hisyam.pdf; http://ir.unimas.my/id/eprint/7814/4/Noor%20Hisyam%20full.pdf; http://ir.unimas.my/id/eprint/7814/

This study proposes the application of Artificial Neural Network (ANN) in the prediction of water level under tidal influence for Sungai Limbang. ANN is undoubtedly a robust tool for forecasting various non-linear hydrologic processes, including the water level prediction. It is a flexible mathematical structure which is capable to generalize patterns in imprecise or noisy and ambiguous input and output data sets. In this study, the ANN is developed specifically to forecast the daily water level for Limbang Station. Distinctive networks were trained and tested using daily data obtained from the Department of Irrigation and Drainage (DID), Samarahan. Various training parameters are considered in order to gain the best prediction possible. The performances of the ANN is evaluated based on the coefficient of efficiency, E2 and the coefficient of correlation, R. Multilayer Perceptron (MLP) and Radial Basis Function (RBF) were adopted in this study. MLP is trained with conjugate gradient algorithms, trainscg and RBF with newrb. The optimal model found in this study is the MLP which is using four days of antecedent data with combination of learning rate and number of neurons in the hidden layer of 0.6 and 60. This model generated the highest E2 and R Testing of 0.950 compared to RBF which gives the highest value of 0.276 for E2 and for R Test is 0.390. It is found that the ANN has the potential to solve the problems of water level prediction. After appropriate simulations, ANN generates satisfactory results for MLP during both of the training and testing phases but not for RBF. Further, strength and limitations of the ANN are discussed, based on the results attained in this study.