Journal of Japan Society of Civil Engineers, Ser. B1 (Hydraulic Engineering)
Volume, Number, Page
Vol. 76
No. 2
I823-I828
Published date
Nov. 2020
Publisher
Japanese:
公益社団法人土木学会
English:
Japan Society of Civil Engineers
Conference name
Japanese:
第65回水工学講演会
English:
The 65th Conference on Hydraulic Engineering
Conference site
Japanese:
オンライン
English:
Online
Abstract
In recent years, there has been a demand for disaster countermeasures that make use of existing infrastructure. In this study, we develop dam operation models using deep reinforcement learning. To investigate
the possibility of pre-discharage operation, we apply this model to pre-discharge operation by considering
inflow prediction and precipitation prediction information. From the result of learning and testing with the
dam operation model, we revealed that we can apply dam operation model to pre-discharge operation. On
the other hand, the action that models select during testing will converge only one action when the predischarge operation and flood control operation are treated as a series of operation. As a result, model can
fail in flood control operation.