Electrical energy consumption estimation by genetic algorithm and analysis of variance

Abstract

This study presents a genetic algorithm (GA) with variable    parameters    to    forecast    electricity    demand    in    agricultural,   low   energy   consuming   and   energy   intensive   sectors  using  stochastic  procedures.  The  economic  indicators  used in this paper are price, value added, number of customers and  consumption  in  the  last  periods  for  agricultural  and  low  energy  consuming  sectors  and  price,  value  added,  number  of  customers,  price  of  the  substitute  fuel  and  energy  intensity  in  energy   intensive   sector.   Three   kinds   of   models;   linear-logarithmic,  exponential  and  quadratic  are  used  to  find  which  leads  us  to  minimum  error  for  the  related  sector.  The  GA  applied  in  this  study  has  been  tuned  for  all  its  parameters  and  the  best  coefficients  with  minimum  error  are  identified,  while  all  parameter  values  are  tested  concurrently.  The  estimation  errors   of   genetic   algorithm   models   are   less   than   that   of   estimated  by  regression  method.  Finally,  analysis  of  variance  (ANOVA)  is  applied  to  compare  genetic  algorithm  (  three  models), regression and actual data.  It is found that at α = 0.05 the  five  treatments  are  not  equal  and  therefore  Duncan  test  is  applied  to  see  which  treatment  pair  has  lead  to  the  rejection  of  null   hypothesis.   Furtherer   more   it   is   shown   that   genetic   algorithm  estimation  is  closer  to  actual  data  with  less  MAPE  (Mean  Absolute  Percentage  Error)  error  than  that  of  estimated  by  regression.  The  data  from  1979  to  2003  is  used  to  forecast  electricity  consumption  in  the  aforementioned  sectors  as  the  case study.