New Artificial Neural Network and Extended Kalman Filter hybrid model of flood prediction system

Accurate prediction of flood water level is a difficult task to achieve due to the nonlinearity of the water level itself and lacking of input parameters to the neural network model. Although Artificial Neural Network is proven to be the best model of flood water level prediction, suitable model parameters need to be chosen for training purposes in order to arrive to an optimal model with smallest error. A new Back Propagation Neural Network model (BPN) for the prediction of flood water level 3 hours ahead of time is developed in this study. This optimized BPN model offers advantages of parameter analysis method instead of trial and error method for choosing the optimized BPN model parameters. However, the simulated results of BPN model required improvement as the model could not able to track the actual water level precisely. Hence, this paper proposes BPN model with integration of EKF at the output. Performance indices result such as Akaike's Final Prediction Error(FPE), Loss Function(V) and Root Mean Square Error (RMSE) from this hybrid model outperform the BPN model result.