Abstract
In this research, we study the multi-agent scheduling problem in flow shop environment. In a multi-agent scheduling problem, the set of jobs is divided into several subsets. There are several agents, each interested in a subset of jobs and has its own objective. In this paper flow shop scheduling problem with two agents is considered, in which the processing sequence of all jobs is to be determined. Each job belongs to either of the agents, the objective function of agent 1 is the makespan of its jobs and agent 2 is the total tardiness of its jobs. Two agents have to compete on the use of common processing resources. Every agent’s objective must be considered in the scheduling of jobs and the final schedule must be acceptable to all agents. Two agents’ objectives are measured in Pareto optimality and final result is a set of Pareto optimal scheduling solutions. A simulation model including information of machines, jobs and agents is constructed to evaluate performances. The search of optimal solutions in problems with large feasible solution space is through Multi-Objective Particle Swarm Optimization. Due to the low convergence of MOPSO, we propose new swarm setting, position updating, personal best position updating and initialization operations based on two-agent features. The improved new algorithm is called Agent-based MOPSO. The parameter analysis is conducted to determine the best parameter combination in searching efficiency and solution quality. In the end, two MOPSO algorithms are compared with Variable Neighborhood Search in two-agent flow shop scheduling problem. Results show that Agent-based MOPSO performs better than the other two algorithms in distance metric, space metric and diversity metric in this problem.