Deep-reinforcement-learning-based optimization for intra-urban epidemic control considering spatiotemporal orderliness | Synapse