Robustness of infrastructure networks with cascading overload failures under extreme geographically localized attacks.

Many infrastructure networks, such as power grids, road networks, and the Internet, are embedded in space. These networks are extremely susceptible to geographically localized attacks caused by natural disasters and will experience cascading overload failures due to node load redistribution. Analyzing the robustness of these spatial networks under extreme geographically localized attacks (EGLAs), in which attacks target the most critical locations of networks, is important for improving their safety. However, existing studies for analyzing the robustness of spatial networks under EGLA only consider the initial failures caused by attacks, ignoring the effect of cascading overload failures. In this paper, we simultaneously consider the effects of the above factors to study the robustness of networks with loads under EGLA. We construct a strategy to identify the locations of the EGLA by maximizing the importance of nodes in the attack regions and propose an adaptive simulated annealing algorithm to solve it. Experiments in the synthetic and real-world networks show that our strategy can find the locations of attacks more accurately compared to traditional methods. Remarkably, we find that the link length distribution of the network can affect the robustness of the network by influencing the betweenness distribution of the nodes. Accordingly, a strategy by randomly adding long-range edges is designed to improve the robustness of real infrastructure networks, which can provide new insights into the safety of systems.