Performance comparison of self-adaptive and adaptive differential evolution algorithms

Notes (per sections)

(1) Introduction

• parameters:
  • parameter tuning -- preselected by the user based on experimentation;
  • parameter control -- selected at run-time;
• parameter control:
  • constant -- the same as tuning;
  • deterministic parameter control -- result of a function;
  • adaptive parameter control -- based on an (automatic, internal) feed-back about performance;
  • self-adaptive parameter control -- the parameter is embedded in the individual, and thus it evolves at the same time with it;

(3) The differential evolution algorithm

• mutation variants:

  • rand/1;

  • best/1;

  • current-to-best/1;

  • best/2;

  • rand/2;

• coercion:
  • reflection;
  • truncate;
  • none;

  • citations: <<Citation(Ronkkonen J, Kukkonen S, Price K (2005))>>, <<Citation(Storn R, Price K (1995))>>, <<Citation(Storn R, Price K (1997))>>;

(4) The adaptive and self-adaptive differential evolution algorithms

• FADE (Fuzzy Adaptive Differential Evolution):

  • adapts only mutation and crossover;
  • population size is fixed by the user;

  • citations: <<Citation(Liu J, Lampien J (2005))>>;

• DESAP:

  • population size:

    • automatically chosen at initialization: ;

    • DESAP-Abs -- ???;

    • DESAP-Rel -- the population is enlarged or shrinked by an uniform factor ;

  • all parameters are self-adapted (including population size???);

  • citations: <<Citation(Teo J (2005))>>;

• SaDE (Self-adaptive Differential Evolution):

  • extensive description of the algorithm;

  • citations: <<Citation(Qin AK, Suganthan PN (2005))>>;

• jDE:

  • the population size is fixed by the user;
  • it self-adapts the crossover and mutation parameters;
  • it embeds those parameters in each individual;
  • adaptation formulas:

    • ;

    • ;

    • ;

  • as coercion truncation (volatile???) is used;

  • citations: <<Citation(Brest J, Greiner S, Boskovic B, Mernik M, Zumer V (2006))>>;

• jDE-2:

  • improved version of jDE;

  • the truncation and reflection coercion is used with equal probability;

  • they use the same strategies as SaDE (rand/1/bin and current-to-best/1/bin);

  • as self-adaptive parameters they use the same scheme as in the case of classical jDE;

  • they have different sets of parameters for each strategy;
  • how do they choose between strategies???;