Flood prediction using NARX neural network and EKF prediction technique: A comparative study

Accurate and reliable flood water level prediction is very difficult to achieve as it is often characterized as chaotic in nature. Prediction using conventional neural network techniques with back propagation algorithm which was widely used does not provide reliable prediction results. Flood water level is characterizing as a dynamic nonlinear properties that cannot be represented by static neural network such as back propagation algorithm. Therefore, NARX NN is propose as the identification model because it could reflect the dynamic characteristics of the flood water level, as NARX structure includes the feedback of the network output. This paper compares the prediction performances of NARX model and EKF prediction technique in flood water level prediction. EKF is well known as the best nonlinear state estimator. Results showed that NARX model performed better than EKF prediction technique.