Prediction Method of Aircraft Stealth Characteristics Based on DNN-Ridge Regression Hybrid Model Driven by Multifidelity Data

With the development of radar stealth technology, aircraft stealth design requires higher electromagnetic calculation. Electromagnetic calculation is the key link of connecting theoretical design, engineering realization and application, and its accuracy and efficiency directly affect the stealth performance evaluation and research and development cycle of aircraft. In order to solve the problem of high calculation cost of high precision radar cross section (RCS) in aircraft stealth design, a new multi-fidelity modeling method is proposed in this paper. The multi-fidelity non-linear mapping relationship is established by DNN, and the loss weight of trough region is given in the loss function, so that the model can strengthen the learning of radar stealth sensitive features. Ridge regression model is used to correct the system deviation of depth learning model, so as to achieve the prediction accuracy of radar stealth performance evaluation. The multi-fidelity modeling method is applied to the stealth characteristics of typical flying-wing layout. We introduce Gaussian process data enhancement strategy to break through the bottleneck of sparse and high-precision data of complex targets and realize the efficient fusion of low-precision big data and high-precision small samples. The results show that the omni-directional RCS prediction results meet the high-precision stealth characteristics of the aircraft, with higher prediction accuracy compared to the traditional Co-Kriging method and faster prediction efficiency compared to high- precision simulation method, which verifies its engineering applicability in complex electromagnetic environment and provides an efficient and reliable research method for aircraft stealth design.