Design of Transmission Networks for Enhanced Resilience under Stochastic Disruption Scenarios using Graph Generative Models

Complex systems such as power networks undergo various operational situations that result in intricate interactions between system components. Although maintaining standard operations of these complex systems is itself a challenge, considering external events that disrupt the normal state and mitigating their damage is a challenging yet crucial task. This paper focuses on a preventive means to improve the resilience of power network designs that can withstand components failures. Power network data was previously collected to form a large data set for training deep learning models that serve as a design generator. With the help of this generative model capable of creating electrical components and network topology, it allows conventional optimization methods to be implemented through its latent space domain. In this study, optimization problems are formulated so that the general performance of the network can be represented through blackout size, redundancy in the generator and power line, and also considers the design components for cost minimization. To subject the system to uncertain disruptive events, random attack and targeted attack scenarios are considered to simulate stochastic failures in the network. Multiple disruption scenarios are applied with the objective of finding a design with the minimum resulting functional loss. The developed design methodology was applied to a benchmark design case study of the IEEE 57-bus transmission network and compared to the original design. The results showed that the developed method is capable of finding network designs with enhanced resilience, especially for scenarios of targeted attacks created based on network centrality.