A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering.

Thesis (M.Eng.) - Asian Institute of Technology, 2000

Hourly flow forecasting is necessary in coastal rivers and estuaries for the purposes of flood control and mitigation, because the water level is greatly affected by the movement of the tidal waves that continually fluctuate and the upland flood discharge as well as local inflows. The interference of the tidal backwater with the upstream flow of the river increases higher flood stages that enlarge flood plain inundation areas and increase potential flood damages. A Neural Network Model known as Back-Propagation algorithm (BP) is presented for forecasting hourly water levels and discharges in the Chao Phraya River at Bangkok Memorial Bridge (Station C4, km 48), Thailand for one, two and three hours in advance. The model considers the river reach from the river mouth at Fort Chula to Bang Sai, a reach length of 112 km. The river flows is influenced by the effects of upstream discharge at Bang Sai (km 112), the tide levels at Fort Chula (km I) and the local inflow along the river reach. The neural network is calibrated and verified based on the observed data and computed results of the MIKE-I I HD model. It was found that the combination of BPNN model and AR(3) model yields even better results of forecasting in this study. However, the comparison of BPNN model with NN model (early stopped by setting target error=0.20)+AR model shows that for the same initial weights and the same output accuracy, the NN+AR model consumes much less computation time especially the node numbers in the neural network model is large.