MOPSO and tradeoff decision strategy for process parameters optimization with limited data

Product quality is vital in industrial production. This paper presents tradeoff decision-based improved multi-objective particle swarm optimization (TD-IMOPSO) for process parameter optimization under limited data. A meta-learning algorithm constructs the process-parameter optimization knowledge model (KMPPO) to define the objective function, avoiding overfitting from sparse industrial data. Multiple global optimization strategies (elite reverse learning, Lévy flight, simulated annealing) enhance MOPSO’s optimization ability and convergence speed. Experiments using standard multi-objective test functions and a steel quality defect dataset demonstrate TD-IMOPSO's superiority over existing multi-objective algorithms, confirming its optimization effectiveness. Compared with the optimal methods MOPSO, NSGA-II, MOEA/D, MOPSO+EPD, and CMODE, the proposed TD-IMOPSO can reduce the average convergence value of the optimal cost function by 45 %, 40 %, 15 %, 43 %, and 10 %.