Artificial Intelligence and Data Science
Online ISSN : 2435-9262
Reinforcement learning for dam operations with an AI agent for autonomous reward optimization
Kimihiro SATO, Kenta HAKOISHI, Masayuki HITOKOTO
Author information
JOURNAL OPEN ACCESS

2026 Volume 7 Issue 3 Pages 441-451

Details
Abstract

To achieve more efficient dam operation, the application of deep reinforcement learning (DRL) to gate operation has been investigated. It is important to balance flood control and water utilization in dam gate operation. However, reward optimization requires iterative trial-and-error based on the analysis of DRL results, and the policy and rationale of reward design are not always clearly documented after the process. As a result, it is difficult to acquire transferable insights for reward design in other d ams. In this study, we propose a framework in which an AI agent automatically improves the reward function of DRL for dam operation and generates natural language reports of the analysis of the improvement process. The framework was applied to virtual flood data for the Ookawa Dam and Miyako water level observation station downstream. The results confirm that the proposed framework can improve dam operation while documenting the reward improvement process.

Content from these authors
© 2026 Japan Society of Civil Engineers
Previous article Next article
feedback
Top