SURROGATE MODELING OF ENGINE PERFORMANCE PARAMETERS USING REGULARIZING GENERATIVE ADVERSARIAL NETWORKS

We propose a new surrogate modeling method based on regularizing generative adversarial models. The method first performs design of experiments (DoE) and numerical experiments on the surrogate object to obtain a basic engine performance dataset. The generated data is used to train an improved regularized generative adversarial network to build a surrogate model of engine performance parameters, solving the difficult problem of high cost and poor security in obtaining reliab濿濸 data in industry. To address the problem that observed data can be corrupted by complex noise in real industrial environments, we add noise to the high-fidelity dataset and verify the feasibility and robustness of the method. Based on the experimental results, our method produces a surrogate model that is able to return predictive outputs that take noise into account. Our method demonstrates superior accuracy compared to other classical surrogate algorithms, successfully capturing the essential characteristics of surrogate objects.