Characterization and identification of electrical customers through the use of self-organizing maps and daily load parameters

This paper shows the capacity of modern computational techniques such as the self-organizing map (SOM) as a methodology to achieve the classification of the electrical customers in a commercial or geographical area. This approach allows to extract the pattern of customer behavior from historic load demand series. Several ways of data analysis from load curves can be used to get different input data to "feed" the neural network. In this work, we propose two methods to improve customer clustering: the use of frequency-based indices and the use of the hourly load curve. Results of a case study developed on a set of different Spanish customers and a comparison between the two approaches proposed here are presented.