Hybrid Approach Based on Simulation and Experimental Data for Surrogate-Based Modeling: A Case Study of a High-Temperature Heat Pump

Kurzfassung

Surrogate models are widely used to approximate complex simulation models, reducing computational costs in real-world applications. However, models based exclusively on simulation data can be inaccurate due to factors like missing real-world effects and discrepancies between simulated and actual performance. Similarly, relying solely on experimental data is costly, time-consuming, and prone to measurement noise and inconsistencies. To overcome these limitations, hybrid modeling has emerged as a promising approach, which integrate both simulation and experimental data, improving the accuracy and reliability of surrogate models. This study investigates methods like transfer learning and adaptive weighting to combine simulation and experimental data effectively, evaluating them in terms of accuracy, robustness, and training time. Specifically, we examine how different hybrid methods perform depending on the amount of available data, dimensionality of the problem, and discrepancy between simulation and experimental data. By testing these techniques on both analytical cases and a real-world high-temperature heat pump, we identify the scenarios in which each method is most effective. Our results demonstrate that hybrid models outperform traditional models based solely on simulation or experimental data, offering improved predictive accuracy and better robustness. These findings highlight the potential of hybrid modeling in real-world industrial applications.