Nowadays the design of complex real electrical, electronic or electromagnetic systems may effectively exploit the characteristics of population based global optimizers. One of the main drawbacks of the adoption of these optimizers in the design of a real system is the difficulty in the introduction, in the optimized design algorithm, of all the heuristic knowledge already available in the field. In order to overcome this problem compact genetic algorithms, classified as estimation of distribution algorithm, could be very effective, since they apply and manipulate a suitable probability vector to represent the distribution of good solutions. Unfortunately, their straightforward implementations usually lack of exploration, and they are easily trapped in local maxima. In order to overcome even this drawback and to develop a compact genetic algorithm with both the required exploitation, of the heuristic knowledge, and the exploration, for avoiding local maxima, in this paper a modified cGA is proposed by implementing more probability vectors and adding a suitable learning scheme to the traditional one in order to ensure the effectiveness of the algorithm. The here proposed new algorithm has been tested on some mathematical test functions and on a typical EM design problem, a microwave microstrip filter synthesis.
Improved compact genetic algorithm for EM complex system design
BUI, VAN HA;ZICH, RICCARDO;MUSSETTA, MARCO;
2012-01-01
Abstract
Nowadays the design of complex real electrical, electronic or electromagnetic systems may effectively exploit the characteristics of population based global optimizers. One of the main drawbacks of the adoption of these optimizers in the design of a real system is the difficulty in the introduction, in the optimized design algorithm, of all the heuristic knowledge already available in the field. In order to overcome this problem compact genetic algorithms, classified as estimation of distribution algorithm, could be very effective, since they apply and manipulate a suitable probability vector to represent the distribution of good solutions. Unfortunately, their straightforward implementations usually lack of exploration, and they are easily trapped in local maxima. In order to overcome even this drawback and to develop a compact genetic algorithm with both the required exploitation, of the heuristic knowledge, and the exploration, for avoiding local maxima, in this paper a modified cGA is proposed by implementing more probability vectors and adding a suitable learning scheme to the traditional one in order to ensure the effectiveness of the algorithm. The here proposed new algorithm has been tested on some mathematical test functions and on a typical EM design problem, a microwave microstrip filter synthesis.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
