Data Mining for Patterns Recognition of Power Systems Static Security Assessment with Contingency Events

Abstract

This  paper  analyzes  the  static  security of  the power system,  applying  advanced  data  mining  techniques that allow the evaluation of safety patterns of a power  electrical   system   in a   steady   state   analysis   with contingency  events  N-1.  Data are obtained  through power  flows,  to  perform  Monte  Carlo  simulations  with    scripts    developed    in    Python.    Using    the  DIgSILENT    PowerFactory    simulation    software, 10,000  scenarios  are analyzed,  which  allows  us  to consider  the  uncertainty  of  the  system  according  to the  probabilistic  nature  of  the  system.  The  static security indexes   of   the   system   are   calculated   to

classify  the  types  of  contingencies  as  safe,  critically safe,  insecure  and  highly  unsafe. Data  mining  is developed by means of an algorithm programmed in  Python   language   with   which   the   design   of   the multiclass  support  vector  machine  classifier  (SVM Multiclass) is carried out. It is trained to determine if a contingency is safe or unsafe. The parameters of the

SVM were  obtained  through  an  optimization with  a differential evolution algorithm (Differential Evolution).   The   results   of   the   validation   of   the classifier  showed  that  the  technique  is  very  effective in classifying new contingencies. The methodology is applied to an IEEE test system of 39 buses.