Integration of time-dependent covariates in recurrent events modelling : application to failures on drinking water networks

Abstract

The standard Linear Extended Yule Process is a well-adapted stochastic model for water pipes failure modelling, as it takes into account the past number of pipe failures, the ageing and the effects of covariates in the failure rate. But fluctuations of failures along time in water network show the limits of the model, which does not consider time-dependent covariates, like frost effect. An improvement of the model actually used in a standard tool such as {\it PREVOIR$\copyright$ Canalisation} is therefore proposed that considers a time dependent covariate. This allows for dramatically improving the predictions of the number of pipe failures according to the climate.