A novel evolution control strategy for surrogate-assisted design optimization

Optimization solutions of real-world engineering problems mainly suffer from the large computational cost, the curse of dimensionality, and the multi-disciplinary nature of the involved disciplines. These issues may be intensified by incorporating uncertainties into the design and optimization of the problem. In this context, Surrogate-Assisted Optimization (SAO) methods and Evolution Control Strategies (ECS) have been considered as powerful paradigms to overcome or at least to alleviate the mentioned issues over the last two decades. This paper presents a novel ECS strategy based on the meta-models along with the real models. This strategy calculates the accuracy of the meta-model at each design point and determines if the real-model needs to be replaced with the meta-model. Moreover, the SAO and ECS are integrated to develop an augmented strategy to solve complex problems like Uncertainty-based Multidisciplinary Design Optimization (UMDO). In this context, the artificial neural networks are used along with the improved Latin hypercube sampling technique. Performance benefits of the proposed strategy in achieving the near-global optimum solution are shown by solving two simple mathematical problems and an engineering benchmark problem. To demonstrate the potential capability of this strategy, it applies to the UMDO problem of a space transportation system. Simulation results illustrate that the proposed strategy improves the computational efficiency as well as the globality of the optimal solution by proper management of the meta-models and real-models within the optimization process.