Forecasting Muar river water quality using radial basis function neural network
Monitoring and analysis of river water quality is an important element in the environmental monitoring policy and management. Fishing, tourism, drinking and most importantly domestic usage require an acceptable level of river water quality. The modeling of complex and nonlinear systems like river is...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2013
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/42127/5/MohdRaziAbdJalalMFKE2013.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Monitoring and analysis of river water quality is an important element in the environmental monitoring policy and management. Fishing, tourism, drinking and most importantly domestic usage require an acceptable level of river water quality. The modeling of complex and nonlinear systems like river is difficult due to the presence of many variables and disturbance. Usually, the dynamic of the problem is modeled using mathematical relationship. However, most of the time a model requires a lot of information and running its simulation needs a significant amount of time. This project attempts to avoid this process by approximating the problem using a type of Artificial Neural Networks (ANN), which is the Radial Basis Function Neural Networks (RBFNN) instead of commonly used ANN: the Multilayer Perceptron (MLP). RBFNN was assessed to forecast water quality in Muar River, Malaysia where historical and lagged data of water quality were used as input for the networks, and forecasting accuracy was evaluated by using Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Correlation Coefficient (CC). It was found that the RBFNN could be used effectively to predict one-day ahead of turbidity and aluminium value of Muar River. The RBF network produced slightly better results in forecasting with lower value of RMSE; 0.0394 and MAE; 0.0208 but higher value of CC; 0.5385 compared to MLP network for value of RMSE; 0.0435, MAE; 0.0230 and CC; 0.5213 in aluminium forecasting. The same observations were also found in turbidity forecasting where RBF network for value of RMSE; 40.3812, MAE; 25.8489 and CC; 0.6821 slightly better than MLP network for value of RMSE; 40.5804, MAE; 26.9558 and CC; 0.6453. RBF network processing time proved to be 77.9% to 80.9% faster than MLP network in forecasting aluminium and turbidity. |
|---|