RUN-OFF PREDICTION METHOD USING SEQUENTIAL LEARNING DISTRIBUTED MODEL

Abstract

A run-off prediction model based on distributed model and sequential learning method has developed. It enable to real-time prediction of the flood and land slide disaster, which are dependent on runoff characteristics. The distributed model, based on the unstructured triangle mesh, is composed of rain infiltration, surface flow, subsurface flow, discharge flow and river flow. To shorten calculation time, we applied parallel computation technique based on MPI. as a sequential learning model, we propose the hybrid method composed of inverse analysis and neural network. We applied this model to the SUMIYOSHI basin, and the predicted result showed relatively good agreement with the observation.