Date of Award
Fall 2006
Document Type
Thesis - Restricted
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
First Advisor
Brown, Ronald H.
Second Advisor
Corliss, George F.
Third Advisor
Povinelli, Richard J.
Abstract
Natural gas is a combustible hydrocarbon composed primarily of methane and accounts for roughly 25% of the total United States energy consumption. There are over 68 million natural gas customers in the US. Natural gas customers are divided into five categories: this thesis focuses on detecting faulty meters for commercial and industrial customers. A natural gas Local Distribution Company (LDC) supplies natural gas to its customers. For this service, the LDC bills the customers at specified intervals. In this study, we detect meter anomalies in the customer data so that the LDC can bill their customers properly. Outlier detection methods such as the 'formula" edit rule and the Hampel identifier are used to identify outliers in the customer data. These outlier detection methods work well for normally distributed data. Outlier detection methods are implemented in mathematical models. We have developed methods to identify anomalies (outliers) in non-normal distributions. In addition, we are applying these concepts to a novel field, natural gas meter readings. We fit the data to a model and look at the residuals for points that are not predicted well by the model. The residual error distribution is asymmetric. That suggests that we apply outlier detection separately to high and low residuals. Outliers are found in the tails, and the tails of our residual distributions are very thick, measured by their kurtoses, compared with a normal distribution. The procedures are applied to an initial data set of 615 meter identifiers (IDs) and generalized to the fulI data set of 3800+ meter IDs. The results are discussed for both outlier detection methods (the 3'1 edit rule and the 'model fitting' exercise) that are used in the study.
Recommended Citation
Kennedy, Rohan O., "Detecting Outliers and Meter Anomalies in Natural-Gas Customer Flow Data" (2006). Master's Theses (1922-2009) Access restricted to Marquette Campus. 4887.
https://epublications.marquette.edu/theses/4887