Development of neural network based rainfall-runoff model for Sungai Pahang, Pekan /
Flooding disaster happens almost annually in Malaysia. With no real solution to this incident, loss of human lives and wealth are inevitable. A forewarning system which could anticipate incoming downpour is required. Heavy rainfall is an important aspect which contributes to flood. Monitoring rainfa...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Kuala Lumpur :
Kulliyyah of Engineering, International Islamic University Malaysia,
2017
|
| Subjects: | |
| Online Access: | Click here to view 1st 24 pages of the thesis. Members can view fulltext at the specified PCs in the library. |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Flooding disaster happens almost annually in Malaysia. With no real solution to this incident, loss of human lives and wealth are inevitable. A forewarning system which could anticipate incoming downpour is required. Heavy rainfall is an important aspect which contributes to flood. Monitoring rainfall remains as an integral part of flood defense system. One of the most leading method to predict flood is by developing a forecast model of rainfall-runoff. Rainfall and river flow relation are very much subjective with various affecting factors. No single model could generally model rainfall-runoff system as climatic, topology and many other factors differ geographically from one place to another. Artificial Neural Network (ANN) is preferred to model hydro system because of its accountability of nonlinear dynamics of water flow. This paper evaluates and compare the performances of an ANN based rainfall-runoff model. A case study is done on a flood-prone river basin in Pahang River, in Pekan. Rainfall data of 5 hydrologic stations and river level are used as the input. The data taken were recorded from the year of 2012 until the end of 2014. Different types of learning algorithms are used to train ANN, namely Levenberg Marquardt (LM), Bayesian Regularization (BR) and Particle Swarm Optimization (PSO). The performance of the learning algorithms involved were evaluated with a coefficient of determination (R) and Mean Square Error (MSE). Network configuration of the ANN which best fit each algorithm is determined. Performance comparison for each learning algorithm in a 12-hour rainfall-runoff forecast is acquired. ANN trained with LM shows average correlation coefficient of 0.996514 in comparison to ANN trained by BR with 0.996186, and ANN trained by PSO with 0.981378. The results achieved showed the promising advantage of LM as compared to BR and PSO. |
|---|---|
| Physical Description: | xv, 83 leaves : illustrations ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 73-77). |