Marjan Ilkovski, prof. Boštjan Blažič

Clustering consumers with similar consumption patterns has many potential applications. Load profiles can be used by energy retailers to improve their business decision making by offering flexible tariffs depending on the season. Other interesting application, where clustering could be applied, is targeting the consumers in demand response [1]. It enables the consumers to play an active role in the competitive electricity markets. A consumer that has accurate information of its consumption pattern is encouraged to alter its consumption in periods, where the generation cost is high. Moreover, load profiles from the clusters can be also effectively combined with forecasting procedures.

Clustering is an unsupervised learning classification technique that enables a division of data into groups of similar objects. Each group, called cluster, consists of objects similar amongst themselves and dissimilar compared to objects of other groups [2]. Clustering consumption refers to the formulation of representative consumption curves for single consumer and groups of consumers. Based on certain criteria, the consumers are grouped together in a number of classes. Each class has a representative consumption curve which is weighted according to the average of the curves that belong to the cluster [3].

There are various types of clustering techniques. In particular, it is possible to identify partitioning methods, such as k-means and k-medoids [3], [4], hierarchical methods [3], [5], [6], density-based methods [7], grid-based methods [8], model-based methods, such as Kohonen’s self-organizing map (SOM) [4], [9], [10].

One example where clustering was used is in a simulation of the operation of distribution networks involving new consumers (electric cars, heat pumps…). The study examined the use of the clustering methodology in metered data of more than 1000 consumers (residential load). The consumers were grouped in clusters, based on k-medoids method, shown on Figure 1. Every consumer was represented with a few indicators (such as skewness, lunch impact, night impact, load factor, low factor etc.), that are related with the shape of the curve. Representatives of each cluster, which are the closest objects to the center of clusters, were then used in simulations.

Figure 1: K-medoids clustering method. k – the number of clusters to be created; A – the method (for initial partitioning) creates partitions, B – the method associates each data point to the closest medoid;
C – checks the objects that are not representative; D – repeats until it has checked all the objects and selects the best representatives.

Figure 2: Consumers were grouped in 20 clusters. The figure shows all profiles in a particular cluster. Black color indicates the representative consumer.

Conclusion

Clustering is an important technique to improve efficiency and sustainability of electricity systems. In modern power system planning, the installation of smart metering in a large portion of the consumers, is a key area of focus. The characteristics of the data stored by the smart meters, and their combination with exogenous variables (meteorological, economical etc.), opens the possibility of designing clustering models for household consumers. The clustering methodology helps us to better understand the behavior of the metered data sets and how this knowledge can be exploited to improve power systems.

References

[1] A. M. Alonso, F. J. Nogales, and C. Ruiz, “Hierarchical Clustering for Smart Meter Electricity Loads based on Quantile Autocovariances,” IEEE Trans. Smart Grid, 2020.
[2] O. A. Abbas, “Comparisons Between Data Clustering Algorithms.,” Int. Arab J. Inf. Technol. IAJIT, vol. 5, no. 3, 2008.
[3] I. P. Panapakidis, M. C. Alexiadis, and G. K. Papagiannis, “Load profiling in the deregulated electricity markets: A review of the applications,” in 2012 9th International Conference on the European Energy Market, 2012, pp. 1–8.
[4] V. Figueiredo, F. Rodrigues, Z. Vale, and J. B. Gouveia, “An electric energy consumer characterization framework based on data mining techniques,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 596–602, 2005.
[5] G. J. Tsekouras, N. D. Hatziargyriou, and E. N. Dialynas, “Two-stage pattern recognition of load curves for classification of electricity customers,” IEEE Trans. Power Syst., vol. 22, no. 3, pp. 1120–1128, 2007.
[6] S. Ramos, Z. Vale, J. Santana, and J. Duarte, “Data mining contributions to characterize MV consumers and to improve the suppliers-consumers settlements,” in 2007 IEEE Power Engineering Society General Meeting, 2007, pp. 1–8.
[7] W. Qiu, F. Zhai, Z. Bao, B. Li, Q. Yang, and Y. Cao, “Clustering approach and characteristic indices for load profiles of customers using data from AMI,” in 2016 China International Conference on Electricity Distribution (CICED), 2016, pp. 1–5.
[8] C.-F. Tsai and S.-C. Huang, “An effective and efficient grid-based data clustering algorithm using intuitive neighbor relationship for data mining,” in 2015 international conference on machine learning and cybernetics (ICMLC), 2015, vol. 2, pp. 478–483.
[9] J. Nazarko and Z. A. Styczynski, “Application of statistical and neural approaches to the daily load profiles modelling in power distribution systems,” in 1999 IEEE transmission and distribution conference (Cat. No. 99CH36333), 1999, vol. 1, pp. 320–325.
[10] G. Chicco, R. Napoli, F. Piglione, P. Postolache, M. Scutariu, and C. Toader, “Load pattern-based classification of electricity customers,” IEEE Trans. Power Syst., vol. 19, no. 2, pp. 1232–1239, 2004.