A Combined Surrogate Model Construction Strategy for Structural Reliability Design

Efficiently constructing accurate surrogate models to approximate expensive limit state functions has always been a key challenge in reliability optimization. As the demands for structural integrity analysis and its optimization design continue to rise, engineering problem-solving solutions reliant on surrogate models are constantly being developed and upgraded. To address this, this study proposes a construction method for surrogate models that consider multiple indicators, to enhance the evaluation efficiency, accuracy, and stability of complex structural systems. By considering the information provided by local variance, the method identifies areas where the surrogate model is deficient and prioritizes sampling in those regions. Stability indicators are incorporated to ensure that the model is not adversely affected by extreme or low-quality data during the training process, thereby improving the quality of the established surrogate model. An entropy weighting approach is employed to combine local variance and stability indicators for calculating the weights of individual surrogate models within the combined model, assisting engineers in mitigating potential risks and enhancing the accuracy of predictions and analyses. Finally, a mathematical example was employed to validate the superiority of the proposed combined strategy, while an engineering example was used to demonstrate the effectiveness of the combined surrogate strategy in optimization design.