Optimization Based on Combining Models of Adaptivebehavior of a Swarm of Agents

A bionic search architecture has been developed to solve the problem of placing VLSI elementsbased on the hybridization of the algorithms of a bee colony and a swarm of chromosomes,which allows you to get out of "local holes" and increases the convergence of the placement algorithm. The initial iterations are implemented by the bee algorithm to provide a broad overview ofthe search area, and the final iterations are implemented by the chromosome swarm algorithm,which ensures the exact localization of the extremum found by the bee algorithm. Agents are representedas a population of chromosomes, which are genotypes for solving the placement problem.The paper describes a modified paradigm of a swarm of chromosomes, which, in contrast to thecanonical method, provides the possibility of searching for solutions in the affine space of positionswith integer values of the parameters. In the search population method of optimization by aswarm of chromosomes, the agents of the population are chromosomes. The chromosome is thegenotype of the optimization object. The essence of the search procedure is the successive changeof the states of the object of optimization (chromosome) by the directed mutation operator and thesearch for the optimal state. An affine-relaxation model (ARM) of a swarm of chromosomes isproposed - this is a graph whose vertices correspond to chromosomes, and arcs correspond toaffine bonds between them. The transition of the chromosome to a new state is carried out using arelaxation procedure. In the work, the directed mutation operator acts as a means of changing thesolution, the essence of which is to change the integer values of genes in the chromosome. Thepurpose of the transition is to reduce the weight of the affine bond between chromosomes. Themechanisms of the directed mutation operator are described. A modified structure of the bee algorithmis proposed. For each base chromosome, a probabilistic choice of a set of chromosomeslocated in the vicinity of the base chromosome is implemented. It is possible to improve the qualityof the developed algorithm by adjusting the values of the control parameters. The time complexityof the algorithm for fixed values of the population size and the number of generations is O(n). Ingeneral, the dependence of the running time of the hybrid algorithm is O(n2) – O(n3).