Water Level Forecasting Using Artificial Neural Network (Ann): A Case Study of Semenyih River, Malaysia

Malaysia is a tropical country that experiences heavy rainfall throughout the year, and is therefore prone to flooding, thus, a sound flood forecasting or warning system is essential. The architecture of a water level forecasting model is important in a reliable flood warning system. In this study, Artificial Neural Network (ANN) modeling is used to forecast the water level for the Semenyih River, by using the data recorded by the Department of Irrigation and Drainage as input and output in the model. In this study, 2 hydrological parameters, namely the streamflow (discharge) and the water level have been chosen as the input and output of the model respectively. The feedforward network is developed as the prediction model and the AutoRegressive network with eXogenous input (NARX) as the forecasting model. In the NARX model, the close loop and multi-step ahead prediction were taken into consideration. The results show that the feedforward network model worked out decently well whilst the NARX forecasting model is only able to produce satisfactory non-future predictions. For the prediction model, with the given input and output data, predicted values of moderate accuracy is obtained. Whereas the NARX model manages to deliver a more outstanding prediction outcome but is unable to simulate future values. The prediction model has an RMSE and a coefficient of correlation value of 0.20658 and 0.78803 whilst the forecasting model has an RMSE value of 0.02752. It is recommended that for future studies, use of more recent, accurate, and consistent data sets, combined with carrying out more modelling types, combinations, or variations, will improve the performance of the model.