This paper addresses the issues of uneven resource allocation and low computational efficiency in the multi-edge task offloading process within manufacturing workshops. A multi-edge task offloading optimization method based on Dynamic Collaborative Particle Swarm Optimization (DCPSO) is proposed. First, to enhance the quality of initial solutions and thereby improve overall optimization efficiency, a hybrid initialization strategy integrating random sampling and fitness-guided greedy mechanisms is designed to achieve a balance between solution diversity and quality. Subsequently, to strengthen the algorithm’s exploration capability in complex spaces, a dynamic subgroup collaboration update mechanism is developed, employing dynamic subgroup partitioning and adaptive particle updates to significantly enhance convergence speed and the quality of offspring solutions. Finally, a mutation mechanism is introduced to augment the algorithm’s local search capability, improving its ability to escape local optima. Experimental results demonstrate that, compared to five baseline algorithms, the DCPSO algorithm exhibits significant advantages in terms of convergence, robustness, and sensitivity.