%A Patricia G?mez Mart?nez
%A Francisco Cubillo
%A Francisco Javier Mart?n Carrasco
%A Luis Garrote de Marcos
%J Water
%T Statistical dependence of pipe breaks on explanatory variables
%X Aging infrastructure is the main challenge currently faced by water suppliers. Estimation
of assets lifetime requires reliable criteria to plan assets repair and renewal strategies. To do so,
pipe break prediction is one of the most important inputs. This paper analyzes the statistical
dependence of pipe breaks on explanatory variables, determining their optimal combination and
quantifying their influence on failure prediction accuracy. A large set of registered data from Madrid
water supply network, managed by Canal de Isabel II, has been filtered, classified and studied.
Several statistical Bayesian models have been built and validated from the available information with
a technique that combines reference periods of time as well as geographical location. Statistical models
of increasing complexity are built from zero up to five explanatory variables following two approaches:
a set of independent variables or a combination of two joint variables plus an additional number
of independent variables. With the aim of finding the variable combination that provides the most
accurate prediction, models are compared following an objective validation procedure based on the
model skill to predict the number of pipe breaks in a large set of geographical locations. As expected,
model performance improves as the number of explanatory variables increases. However, the rate
of improvement is not constant. Performance metrics improve significantly up to three variables,
but the tendency is softened for higher order models, especially in trunk mains where performance is
reduced. Slight differences are found between trunk mains and distribution lines when selecting the
most influent variables and models.
%N 3
%K pipe breaks; explanatory variables; predictive models; statistical dependence; distribution
lines; trunk mains; water supply
%P 1-24
%V 9
%D 2017
%I MDPI
%R 10.3390/w9030158
%L upm46389