Ant colony optimisation for coloured travelling salesman problem by multi‐task learning

Abstract

Traditional algorithms, such as genetic algorithm and simulated annealing, have greatly attracted a lot of research studies due to their simplicity and flexibility to solve coloured travelling salesman problem (CTSP). However, their performance is limited in solution quality and convergence speed. To improve these limitations, this study presents a fast and effective ant colony optimisation (FEACO) algorithm. In the proposed FEACO, a new pheromone updating mechanism is incorporated into the traditional ant colony optimisation (ACO) to improve its performance. Furthermore, a multi-task cooperative learning approach is employed to solve CTSP. In it, multiple tasks in a shared city set are cooperatively carried out by all salesmen, but each task in exclusive city sets is independently performed by an appointed salesman. In ACO and FEACO, feasible solutions can be found for CTSP by cooperative learning, which is carried out by the cooperation of a set of ants. During the process, ants can cooperate to find good solutions by utilising the pheromone deposited on the paths while ants pass them. Compared with state-of-the-art algorithms, the experimental results on both small scale and larger scale show that the proposed method has potential to provide an improvement in solution quality and convergence speed.