Discovering patterns in electricity price using clustering techniques

Abstract

Clustering   is   a   process   of   grouping   similar   elements  gathered  or  occurred  closely  together.  This  paper  presents  two  clustering  techniques,  K-means  and  Fuzzy  C-means, for the analysis of the electricity prices time series. Both algorithms  are  focused  on  extracting  useful  information  from  the  data  with  the  aim  of  model  the  time  series  behaviour  and  find   patterns   to   improve   the   price   forecasting.   The   main   objective,  thus,  is  to  find  a  representation  that  preserves  the  original  information  and  describes  the  shape  of  the  time  series  data  as  accurately  as  possible.  This  research  demonstrates  that  the  application  of  clustering  techniques  is  effective  in  order  to  distinguish  several  kinds  of  days.  To  be  precise,  two  major  groups  can  be  distinguished  thanks  to  the  clustering:  the  first  one  that  includes  the  working  days  and  the  second  one  that  includes  weekends  and  festivities.  Equally  remarkable  is  the  similarity shown among days belonging to a same season.