Hybrid metaheuristics can effectively tackle multi-objective optimization problems. Recently, researchers gained interest in procedures, referred to as architectures, that can provide generic functionalities and features for hybridizing arbitrary metaheuristics. Although a previously proposed multi-agent architecture, MO-MAHM, achieved high-quality solutions for bi-objective problems, its application for more than two objectives requires further discussions. To this end, MO- \(\hbox {MAHM}_E\) , a MO-MAHM extension for handling two or more objectives, is proposed in this paper. MO- \(\hbox {MAHM}_E\) maintains concepts from particle swarm optimization and multi-agent paradigms, including particle movement and agent intelligence. Further, it uses a decomposition-based velocity operator prescinding from aggregation functions and reinforcement learning automata scheme for supporting the decisions of the agents. This paper shows that these algorithmic components can significantly improve the architectural performance. We apply MO- \(\hbox {MAHM}_E\) to the quadratic assignment problem with up to four objectives. Hybridization combines three evolutionary algorithms and a local search. A comparison of the experimental results of MO- \(\hbox {MAHM}_E\) and ten algorithms (including hybrid approaches, hyper-heuristics, and algorithms from the quadratic assignment problem literature) is presented.
