Knowledge sharing in cooperative compact genetic algorithm

Abstract

The compact genetic algorithm is derived from the genetic algorithm in which the population is represented by the probabilistic vector. To improve the search capability and avoiding local minima, parallelization has been employed where many search processes are deployed concurrently. In order to coordinate the work of multiple processes, knowledge sharing is necessary. Multiple processes share their probabilistic vectors partially. To escape from local minima the restart step is introduced. The experiment compares the proposed algorithm with two other competitive algorithms using Traveling Salesman problem, Bin Packing problem, Subset Sum problem, and Knapsack problem. The results show that the proposed algorithm is more efficient in finding solutions than the competing algorithms. The detailed analysis of the restart step provides insight into the behaviour of the proposed algorithm.