Modeling of Real-time Prediction with AI Technologies for Distributed Runoff Model

Abstract

As the serious floods due to typhoon or concentrated heavy rainfall have been happened at many areas in the world, the total control system for the whole river basin is getting significant. The distributed runoff model can be provided through newly arranged mesh data and GIS technologies. From the viewpoint of real-time flood prediction to prepare the necessary countermeasures or actions, the system should keep the quick response and high accuracy against abnormal rainfall events with new methodologies.In this research artificial intelligence is used to develop a distributed flood forecast system capable of real-time simulation of river flood levels at all locations within a watershed. Parameter calibration of a distributed rainfall-runoff model (Hydro-BEAM) is carried out using a particle filter, and real-time modeling of river discharge is enabled through the use of a State-Space Neural Network. Optimal training of the neural network is carried out using Optimal Brain Damage, which systematically reduces weak links in the network to allow for improved forecast efficiency and accuracy. An application is made for the Nagara River watershed to demonstrate the effectiveness of the system.