For large customers, Electricite de France - the French national electric power company - stores every 10' the amount of electric power they consume. For each customer, these measures lead to curves called electric load curves. Clustering of electric load curves is a key problem for understanding the behavior of these customers. Several methods have been used but Kohonen maps give a very nice solution to this problem thanks to the visualization of the map. The work we present here describes a software for interactive construction and interpretation of a Kohonen map clustering, in the case of curves. The user can run the Kohonen map clustering, visualize the map, see external characteristics of curves linked to each cell of the map, find the cells figuring curves having some chosen external characteristics, define classes of cells, add comments on cells. User interaction is largely based on mouse clicking on the map cells and on bars of barcharts figuring external characteristics of the curves. This software is not dedicated to electric load curve analysis but can be used on any type of curve, for instance to analyze time series in finance.
[1]
Teuvo Kohonen,et al.
Self-Organizing Maps
,
2010
.
[2]
Georges Hébrail,et al.
Interactive Interpretation of Hierarchical Clustering
,
1997,
Intell. Data Anal..
[3]
Ramakrishnan Srikant,et al.
The Quest Data Mining System
,
1996,
KDD.
[4]
Gilbert Saporta,et al.
Probabilités, Analyse des données et statistique
,
1991
.
[5]
Yeuvo Jphonen,et al.
Self-Organizing Maps
,
1995
.
[6]
C. Muller,et al.
Visualizing 2665 electric power load curves an a single A4 sheet of paper
,
1996,
Proceedings of International Conference on Intelligent System Application to Power Systems.
[7]
Eamonn J. Keogh,et al.
A Probabilistic Approach to Fast Pattern Matching in Time Series Databases
,
1997,
KDD.