In determining loading patterns for use in a transient gas flow model, a cluster analysis computer application has been used at Union Gas for a good number of years. The application was put in place following a presentation at the 1975 Annual Meeting of PSlG (Application of Cluster Analysis in Determining Transient Flow Loading Patterns, by Howard M. Boyer, Southern California Gas Company). A recent revision of the application undertaken at Union Gas takes advantage of much greater computing capabilities than those 10–15 years ago. It is suggested that specific requirements in the application of the cluster analysis method have to be followed. Characteristics of the load points, to which cluster analysis is applicable, and some limitations of the method have been identified
Following the presentation by Howard M. Boyer [1] at the PSlG meeting in 1975, the cluster analysis FORTRAN program developed by him was implemented at Union Gas. It has been used since then in the annual task of determining the hourly loading patterns for Union's transmission systems. Due to rapid growth in the level of gas transportation at Union, the computer program and the procedures of applying it have recently undergone a review and revisions. The purpose of this review was to better understand the implications of the program and to be in a stronger position to defend before regulatory bodies the facilities planning decisions based on the clustering of load profiles. The presentation includes two major sections: the methodology and the application of cluster analysis. The methodology section emphasizes the need to analyze all load points together in a coordinated manner; points out the non-uniqueness of the clustering solution; and ties the output of clustering with the purposes of the transient flow modeling. The application section presents the place of cluster analysis in the transmission planning function at Union Gas and outlines its role in the process of preparing input data for a transient gas flow model. It also illustrates the sensitivity of the transient model capacity to the use of varying profile patterns, and the conditions which produce these patterns. ed.
Cluster Analysis does not appear to be a classical, well established branch of Statistics. There are not many textbooks or monographs with a chapter on this topic. We can point out two different statistical connotations of the term ‘Cluster’. One concept [2] (numbers in brackets are those in the section References) presents Cluster Sampling as a method of the Survey Sampling techniques. The underlying idea is the cost saving when conducting a survey. Let us suppose the entire population space is divided into a number of areas (clusters). The method involves randomly selecting a subset of m clusters from the full set of M clusters, and then conducting the survey analysis for each of the selected clusters. Under this definition, a judgemental division of the population into clusters precedes a statistical analysis within selected individual clusters.