Ense mble of M etamodels with Optimized Weight F actors

摘要: Approximate mathematical models (metamodels) are often used as surrogates for more computationally intensive simulations. The common practice is to construct multiple metamodels based on a training data set, evaluate their accuracy, and then use only single model perceived the best while discarding rest. This has some shortcomings i t does not take full advantage of resources devoted constructing different metamodels, it assumption that changes in set will jeopardize accuracy selected model. It possible overcome these drawbacks improve prediction surrogate if separate stand -alone combined form an ensemble. Motivated by previous research committee neural networks ensemble models, technique developing accurate presented this paper. Here, selection weight factors general weighted sum formulation treated optimization problem with desired solution being one minimizes error metric. proposed evaluated considering industrial four benchmark problems. effect metrics estimating at either or few validation p oints also explored. results show optimized provides predictions than most problems even surpassing ly reported approaches .