Coevolutionary Binary Ant Colony Algorithm

Ant Colony Optimization, firstly introduced in 1990s, is a algorithm of bionics. In the early stages of development, Ant Colony Optimization algorithm is applied to discrete combination problems, but with the development of the algorithm, Ant Colony Optimization algorithm is also gradually applied to continuous domains.This thesis gets: the smallest iterate times of reaching the lower bound of pheromones that is not belongs to solution component of best-so-far solution. And the smallest iterate times of getting to convergence is obtained, then the minimum probability of finding no better solution of iterate t--LPB(t)is figured out. According to this minimum probability and letting the algorithm get best search ability, search bias controlling function is designed. Based on this function, binary ant colony algorithm with controllable search bias is proposed. Further more, after analysis of how parameters, vists and pheromone relate to the performance of the algorithm, two kind of method of parameter setting are designed: parameter setting of known search bias and parameter setting of unknown search bias. The experiment results show its strong ability to locate good solutions and fast convergence speed.According to the self-organization, self similarity and attractor of complex network, a new kind of co-evolutionary algorithm—co-evolutionary binary ant colony algorithm is proposed. The algorithm regards each ant population as a node of complex network, each population compete or cooperate with its neibor for improving own fitness, and through the small word phenomenon of complex network, the effect of cooperation and competition can diffuse to the whole network. And each population also improve own fitness by study. Tested by multiple different multi-modal problems, the experiment results show strong ability to get multi-peaks.Finally, we use the binary ant colony algorithm with controllable search bias to solve Binary quadratic programming problem, Through the experiment of ten standard test problems provided by ORLib, the algorithm could get all optimal solution, exhibits great stable search ability.