A unified ensemble of surrogates with global and local measures for global metamodelling

Surrogate models are widely used in engineering design and optimization to substitute computationally expensive simulations for efficient approximation of system behaviours. However, since actual system behaviours are usually not known a priori, it is very challenging to select the most appropriate surrogate model for a specific application. To tackle this, ensemble models that combine different surrogate models have been developed based on global measures and local measures respectively. This article proposes a novel ensemble of surrogates to take advantage of both global and local measures, and a unified strategy is conceived over the entire design space with proper trade-off between these two measures. The effectiveness of the proposed model is tested with 38 mathematical problems and an engineering optimization example. It is concluded that the proposed model has superior accuracy while keeping comparable robustness and efficiency with other ensemble models. The proposed model is also extended to non-uniform experimental design.z