SIMULATION MODEL OF A CONTROLLED GENETIC ALGORITHM BASED ON PETRI NETS

Abstract

When the genetic algorithm is operating while solving the problems in various subject areas, the problem of changing the order of actuation and parameters of the algorithm operator functioning appears. It is related with the fact that in operation the evolution procedure may come across the population hitting the local extremum, fading, etc. The way out of the current situation can be the increase of damaging capability of the operator. The damaging capability can be increased by changing the parameters of operator functioning, types of operators and actuation order. When high damaging capability of operators is applied, situations are possible when the time for solution search is considerably increased, since dispersion of values of the target function for population individuals grows up significantly. In this case, it is reasonable to decrease the damaging capability thus leading to a more detailed investigation of the search space. Since this procedure is required to be performed directly in the process of solution search, the development of new models and methods of genetic algorithm control is needed. In this work it is proposed to apply the neural network approach to solving the problem of evolution procedure control. For this purpose it is required to carry out adaptation of both the genetic algorithm and artificial neural networks. It is proposed to apply the Petri net theory as the main mathematical apparatus to describe the controlled genetic algorithm. The paper presents the imitation model of the evolution procedure based on the Petri net theory that allows for performing the variation of parameters of operator functioning during the search of solutions.