Classification of infrastructure networks by neighborhood degree distribution

TL;DR

This paper introduces the neighborhood degree concept to better classify infrastructure networks, demonstrating that it provides a more reliable probabilistic model fit, especially with water distribution networks, compared to traditional nodal degree analysis.

Contribution

The paper proposes using neighborhood degree instead of nodal degree for classifying infrastructure networks, enabling more reliable probabilistic modeling.

Findings

Neighborhood degree spans a wider range than nodal degree.

Poisson distribution fits well for neighborhood degree in water networks.

Standard nodal degree classification is less reliable for spatially constrained networks.

Abstract

A common way of classifying network connectivity is the association of the nodal degree distribution to specific probability distribution models. During the last decades, researchers classified many networks using the Poisson or Pareto distributions. Urban infrastructures, like transportation (railways, roads, etc.) and distribution (gas, water, energy, etc.) systems, are peculiar networks strongly constrained by spatial characteristics of the environment where they are constructed. Consequently, the nodal degree of such networks spans very small ranges not allowing a reliable classification using the nodal degree distribution. In order to overcome this problem, we here (i) define the neighborhood degree, equal to the sum of the nodal degrees of the nearest topological neighbors, the adjacent nodes and (ii) propose to use neighborhood degree to classify infrastructure networks. Such neighborhood degree spans a wider range of degrees than the standard one allowing inferring the probabilistic model in a more reliable way, from a statistical standpoint. In order to test our proposal, we here analyze twenty-two real water distribution networks, built in different environments, demonstrating that the Poisson distribution generally models very well their neighborhood degree distributions. This result seems consistent with the less reliable classification achievable with the scarce information using the standard nodal degree distribution.