RESEARCH ON PREDICTION METHODS FOR DAM INFLOW WITH LEAD TIME OF 24 HOURS APPLICABLE TO NO-EXPERIENCE FLOOD EVENTS USING ELASTIC NET

Abstract

 This study proposes dam inflow prediction methods that can be used for no-experience flood events. The frequency of large-scale flooding has been increasing in Japan, and to minimize damages, reliable dam inflow prediction is necessary to implement operations such as pre-emptive discharge.

 In this study, Elastic Net, a sparse modeling method, was used to identify intra-data relationships from limited information in order to predict dam inflow from 24 h of related climate and water data. The results showed that even in cases such as the August 2016 downpour in Hokkaido when heavy rains continued for some time and soil dampness changed, a conservative prediction is possible by adding the effective rainfall with a half-life of 720 h to the explanatory variables. Finally, the study confirmed that predictions suitable for practical use could be obtained for multiple cases of no-experience flood events.