Learning-Based Multi-agent System for Solving Combinatorial Optimization Problems: A New Architecture

Abstract

Solving combinatorial optimization problems is an important challenge in all engineering applications. Researchers have been extensively solving these problems using evolutionary computations. This paper introduces a novel learning-based multi-agent system (LBMAS) in which all agents cooperate by acting on a common population and a two-stage archive containing promising fitness-based and positional-based solutions found so far. Metaheuristics as agents perform their own method individually and then share their outcomes. This way, even though individual performance may be low, collaboration of metaheuristics leads the system to reach high performance. In this system, solutions are modified by all running metaheuristics and the system learns gradually how promising metaheuristics are, in order to apply them based on their effectiveness. Finally, the performance of LBMAS is experimentally evaluated on Multiprocessor Scheduling Problem (MSP) which is an outstanding combinatorial optimization problem. Obtained results in comparison to well-known competitors show that our multi-agent system achieves better results in reasonable running times.