A Preference-based Method of Updating the Surrogate Model by Broad Learning and Its Application

A surrogate model is normally employed to evaluate an individual instead of time-consuming and expensive simulation. Inaccuracy model learning by the limited samples may mislead the evolution process. Thus, effectively updating the model by learning more typical samples is necessary. In this paper, a preference-based method of updating the surrogate model by broad learning is proposed, with the purpose of improving the accuracy of the model with the least computation complexity. The preferred individuals with the least distances to the Pareto front are chosen as the infilling samples. Following that, the surrogate model is updated based on the infilling samples by broad learning, so as to reduce the learning time. The rationality of the proposed updating method is verified by the instance of constructing a surrogate model for bolt supporting quality. The experimental results show that the preference-based updating method has the best accuracy and the least running time. The obtained optimal solutions meet the actual preference of the decision makers with the best convergence.