Deep sparse autoencoders-based community detection and resilience analysis of interdependent infrastructure networks

This paper considers the global information of nodes and constructs a similarity matrix based on s-hop counts. It effectively extracts low-dimensional feature matrices from high-dimensional data to achieve community detection results by utilizing deep learning techniques and deep sparse autoencoders. We successfully detect communities and identify critical inter-community edges. Additionally, we delve into the influence of vulnerable inter-community edges on the resilience of interdependent networks. To illustrate this, a widely employed artificial interdependent power-communication network is adopted as a case study, examining various failure intensities and coupling modes. This approach allows visualization communities, and the impact of vulnerable edges on the interdependent network's resilience is investigated from both structural and functional perspectives. Results have shown that damage to edges bridging different communities can lead to severe network vulnerability. Accordingly, prioritizing the security of these edges will strengthen the network's resilience, which is crucial for preventing further network damage.