Demand Estimation with Automated Meter Reading in a Distribution Network

Accurate estimation and prediction of the demand patterns of the customers of a water works would enable more accurate network hydraulic management. In this study, a probabilistic model to generate residential demand patterns for single-family and semidetached houses is formed. To form these pattern models, an automated meter reading technique has been utilized to gather the necessary information from a sample of residences, which are used to model the demand behavior in a way that is applicable to a much wider range of residences. A linear regression model was constructed to predict the measured average weekly consumption from the calculated average weekly consumption. The residences were clustered by their weekly water demand into four distinct classes using the k-means algorithm. For the final result, probability models developed on the basis of mixtures of Gaussians for each class, in conjunction with the prediction model of the weekly water consumption, were utilized so that estimates for the demand pattern are obtained by sampling the probability distributions for individual single-family and semidetached houses. DOI: 10.1061/(ASCE)WR.1943-5452.0000131. (C) 2011 American Society of Civil Engineers.