A high-fidelity simulation model calibration method based on sequential sampling surrogate model

All models are approximations of the real physical world. To enhance simulation accuracy, parameters must be calibrated using measured data. However, high-fidelity simulation calibration faces prohibitive computational costs due to the inherent complexity of model calculations and the extensive use of simulations during the calibration process. To address this computational challenge, this paper proposes a calibration method based on sequential sampling surrogate models. By replacing high-fidelity simulations with surrogate models and employing a sequential point-adding strategy, the method progressively approaches optimal parameter values, achieving the best possible parameter settings with minimal simulation calls to minimize prediction-measurement errors. This approach is applicable to multi-disciplinary simulation models in mechanics, thermal analysis, and fluid dynamics, demonstrating strong engineering applicability and broad implementation potential. A spacecraft thermal analysis model calibration case validated the method's effectiveness: Through 100 high-fidelity simulation calls, five model parameters were calibrated with errors below 5%, resulting in an average prediction-measurement error of <0.5°C.