An Adaptive Ensemble Surrogate Model Based on Fuzzy C-Means and Its Parallel Infilling Strategy

Abstract To enhance test accuracy and address low modeling efficiency in the practical engineering applications of adaptive ensemble surrogate models (AESMs), this article improves the modeling method for AESMs and its individual adaptive infilling (IAI) strategy. An adaptive ensemble surrogate model based on fuzzy C-means (FCM-AESM) is initially proposed. The global accuracy of AESMs is guaranteed by means of applying FCM analysis to the initial training set, partitioning the training set into chunks, and sieving from the model library. And via standard testing functions, it is validated that the FCM-AESM demonstrates a superior model prediction performance. Subsequently, to further enhance the efficiency and accuracy, a parallel adaptive infilling (PAI) strategy based on the ensemble surrogate model (ES-PAI) is proposed in combination with the IAI strategy. The strategy optimizes new samplings and eliminates those points in the design domain that are in close proximity based on the Euclidean distance criterion, thereby ensuring a uniform distribution of sample points. The influence of employing diverse IAI strategies within the ES-PAI is investigated, along with the examination of whether this strategy can be applicable to the majority of the presently available ensemble surrogate models (ESMs). The outcomes reveal that the ES-PAI strategy consistently surpasses the IAI strategies in both global and local performance and exhibits greater robustness. Eventually implemented in the multiobjective optimization (MOOP) of a forklift gantry, the FCM-AESM method assisted by ES-PAI reduced the gantry's weight by approximately 22.18% while satisfying stress and deformation constraints.