Wyniki wyszukiwania - BazTech

Ograniczanie wyników

2 Prace Naukowe Politechniki Warszawskiej. Elektronika

Znaleziono wyników: 2

Liczba wyników na stronie

Wyniki wyszukiwania

Sortuj według:

Ogranicz wyniki do:

A Comparative Study of Various Strategies in Differential Evolution

Boryczka U., Juszczuk P., Kłosowicz L.

Prace Naukowe Politechniki Warszawskiej. Elektronika

2009

z. 169

19-26

This paper presents a comparison of various strategies of differential evolution. Differential evolution (DE) is a simple and powerful optimization method, which is mainly applied to numerical optimization and many other problems (for example: neural network train, filter design or image analysis). The comparison of various modifications (named strategies) of DE algorithm allows to choose the algorithm version which is best adjusted to desirable requirements. Three parameters are tested: speed, accuracy and completeness. The first section of this article presents general optimization problem and says a little about methods used to function optimization. The next section describes differential evolution - basic algorithm is presented. Two different crossover methods, process of initial population creation and basic mutation schema are described. The third section describes the most popular DE strategies. In the fourth section a new modification (called λ-modification) of DE algorithm is presented. Next section provides basic information about four test functions and differential evolution parameters used in research. The paper presents then summary and final conclusions.

Initialization with repetitions method for sequential niche techniques

Boryczka U., Kłosowicz L.

Prace Naukowe Politechniki Warszawskiej. Elektronika

2008

z. 165

51-58

This paper presents a new method of initialization population for sequential niching techniques, in which, evolutionary algorithm (EA) for determining a single local extreme has been employed. Knowledge of the localization of optima, determined in earlier runs of EA has been exploited in this approach. Initialization of a single individual consists in repeating its location until it is placed in the search subspace does not connected with any niche determined earlier. This approach contributes to the enhancement of convergence and to the improvement of achieved results.