Efficient network resilience computation by preserving social impact estimates

High-fidelity modeling of population, sectors, or goods is required for impact assessment of infrastructure systems, particularly to distribute resilience investments.Two common challenges arise in such a context.First, aggregated metrics at the system level mask the impact of service disruption on potentially vulnerable subpopulations, economic sectors, or types of goods.Second, estimating networked systems' performance over high-fidelity models becomes computationally expensive in probabilistic assessments.This paper proposes the use of attributed networks to improve the formulation and analysis of community resilience.Herein, 'users' are assigned to the physical network while preserving their labels and associated weights.Using the new user-physical model, we propose the extension of connectivity metrics from their network-only version to a user-weighted one, and to find 'user-physical communities', which helps uncover hidden interactions between the physical and social systems.We further use the 'communities' to reduce computational demands while keeping intact the infrastructure model.To illustrate, we use computational experiments on hypothetical infrastructure and social systems.Comparisons of the attributed-graph performance estimates with their equivalent unweighted (traditional) versions show that the former avoids the 'masking' problem.Also, the proposed metrics showed that while fidelity reduction using network-only approaches can bias decision-making for some labels (e.g., a social group), the community-based reduction conserves the information of user-physical communities, thus the expected performance per user type.By enabling the quantification of subpopulation impacts, this contribution advocates for more equitable estimates while guiding resilience investments.