CONSTRUCTION OF DAM-OPERATION-MODEL CONSIDERING DOWNSTREAM AREA OF DAM USING DEEP-REINFORCEMENT-LEARNING

Abstract

 We developed the deep neural network model of dam operation(AI model) for flood control considering the flood risk at the downstream. First, we created a regression equation that predict the downstream river-discharge using the dam discharge as input data. We made reward function of the reinforcement learning which consider the difference between capacity of river channel and estimated river discharge.

 AI model was trained through the proccess of trial and error to obtain dam operation to reduce the peak discharge at the downstream station. The result of AI model was compared with the operation rules and previous AI model which does not consider the flood risk at the downstream, and consequently validity was confirmed.