Optimisation of waterlogging disaster warning system based on deep reinforcement learning

This study focuses on the optimisation of the urban waterlogging disaster warning system. In view of the limitations of traditional warning systems in complex urban environments, deep reinforcement learning (DRL) technology is introduced. The proposed DRL-EWS model integrates a spatiotemporal graph convolutional network for spatial feature extraction, a gated recurrent unit for temporal sequence modelling, and a policy gradient-based decision module for adaptive warning generation. By innovatively combining these components, the risk of urban waterlogging is accurately predicted. The results show that the waterlogging disaster warning system (DRL-EWS) based on DRL is superior to traditional and other machine learning warning models in terms of warning accuracy, false alarm rate, and false alarm rate, and has good adaptability to different terrains, pipe network density, rainfall patterns, and other conditions. This study not only enriches the waterlogging warning technology system in theory, but its practical results are expected to enhance the urban waterlogging defence capabilities and provide a strong guarantee for urban safety. At the same time, it also points out the direction for further improvement in subsequent research.