Deep Learning-based Critical Infrastructure Simulation Model for Disaster Monitoring

In this paper, we describe a Deep learning-based approach for Real-time Critical Infrastructure Protection in the scenario of a Flood event. This would help us understand the dependencies among various Critical Infrastructures and the severity of the current situation. We propose a Multiagent Deep Reinforcement Learning technique to design the policy and reward functions, which the agent must follow. Further, integrating Reinforcement Learning with GIS aids in putting the model in a spatial context and multiple-layer visualization leading to enhanced awareness of the situation. It also helps in the understanding of the spatiotemporal relationships among them. Each of the Geospatial agents will have its state and a set of actions that it needs to take. The agents will act with respect to their dependence on other related Infrastructures and take the best possible action as the disaster unfolds so that immediate response can reduce the severity of the damage. Real-time information simulation would help disaster response personnel to develop various scenarios in the simulation environment and see how the set of critical infrastructures are responding over time as the severity of the flood increases.