Abstract
As the frequency of droughts increases with climate change, more effective use of existing dams is required. In this study, we developed a model to support the optimal operation to effectively use the capacity of dams using reinforcement learning and verified its usefulness. The developed model predicts the optimum replenishment quantity based on various conditions such as the water storage quantity and inflow quantity of each dam at the present time. In the reinforcement learning model, it was possible to predict the shortage of the reference point and the operation with less invalid discharge compared with the operation in which the replenishment quantity is set according to the water storage ratio of each dam. In addition, as a method for effectively utilizing the capacity of a group of dams, efficient operation in accordance with the water storage conditions is required based on the characteristics of individual dams, such as the inflow volume and storage volume. The predicted amount of supplementation by the reinforcement learning model was almost consistent with this trend.
