1 Abstract—Classifying consumers, namely LV consumers, in order to assign them typical load diagrams, was always a concern of the electric utilities, which used this kind of information to better manage their distribution networks. Now, with the transition to a completely open market, the need for settlement between distribution operators and traders requires hourly consumption records that are not generally available, so deriving load diagrams for LV consumers is a mandatory task. This paper presents a new methodology for this purpose that uses typical diagrams obtained in measurement campaigns to create classes defined in the commercial information space that maximize the compactness of the diagrams in each class. The methodology was developed in a project with EDP (the Portuguese distribution operator) and the result will probably be adopted by the regulatory authority. Index Terms-- Clustering methods, Load modeling, Electricity Markets, Energy Management, Optimization methods, Simulated annealing I. INTRODUCTION LASSIFYING consumers, namely LV consumers, in order to assign them typical load diagrams, was always a concern of the electric utilities, which used this kind of information to better manage their distribution networks. Now, with the transition to a completely open market, the need for settlement between distribution operators and traders requires hourly consumption records that are not generally available, so deriving load diagrams for LV consumers is a mandatory task (the expensive alternative is to install interval meters in all LV consumers). The problem of assigning diagrams to consumers, or load profiling, includes a basic dilemma that comes after a first phase of measurement campaigns to obtain a sample of diagrams: either you pre-define the classes, on the basis of commercial information you have for all customers (hired power, energy consumption, activity, etc.) or you cluster a sample of diagrams. In the first option, the diversity of diagrams in each class may be excessive, so the average diagrams are far from being the prototypes even of the sampled diagrams. In the second hypothesis, you get good prototypes, but you are unable to relate them to the commercial information, so you can not classify the consumers outside the sample. The scientific literature on power systems includes a large variety of clustering methodologies applied to the load profiling question [1-14]. Most of the reported techniques fall into one of the two precedent categories, relatively to both the definition of typical load diagrams and the classification process. However, to the knowledge of authors, there’s no standard established method, of relating a class profile obtained through clustering with the consumers’ commercial characteristics. That’s why most electrical distribution utilities used an a priori classification of consumers based on its commercial characteristics. This paper presents a new methodology that uses typical diagrams obtained in measurement campaigns to create classes (defined in the commercial information space) that maximize the compactness of the diagrams in each class. The basic idea is to create first small classification cells in the commercial information space that are afterwards aggregated by an optimization procedure that minimizes a measure of compactness of the diagrams in each class. The methodology was developed in a project with EDP (the Portuguese distribution operator) and the result will probably be adopted by the regulatory authority. The structure of the paper is the following. In the next section, the initial step of creating the basic classification cells is described. Then in Section III, we present the clustering methodology and the optimization process behind it, followed (section IV) by the explanation of the process of applying the class diagram to a specific costumer. Finally, in section V, we show partial results of the project carried out in Portugal. The conclusions and references complete the paper. II. DEFINING THE BASIC CLASSIFICATION CELLS In order to apply the methodology, two blocks of data must be available: - Commercial data of all the LV customers (hired power, energy consump tion, etc); - A sample of load diagrams for the period under analysis (e.g. one year), linked with the same kind of commercial information. The first block of data is generally included in the utility’s databases and raises no problems, except for the need of processing a large number of records. On the other hand, gathering the sample diagrams requires the implementation of a measurement campaign that may be Deriving LV load diagrams for market purposes using commercial information M. A. Matos, Member, IEEE, J. N. Fidalgo, Member, IEEE, L.F. Ribeiro C