Application of Hybrid Physical Electromagnetic Modeling Based on GRNN-XGBoost in Nonlinear Complex Parts of Large End-Winding

In order to better combine the knowledge of physics-informed model and data-driven, and transfer the test model to the real unit-model in the “Smart Turbine”, the numerical reliable simulation data will be further realized, and the real operation data will be mutually verified and the hybrid modeling will be driven. 600 MW large turbo-generator is taken as the research object, and the mathematical and physics-informed model of electromagnetic force for the stator-end winding is established based on the method of image and the air-gap circular current, applying superposition, the Biot-Savart law and the Ampere law. And based on the electromagnetic model of finite element of generator-end, the comparison model of finite element of electromagnetic force for the stator-end winding is established by using external-current loading-excitation source and vector magnetic-potential method (“A” method). At the same time, a hybrid modeling, electromagnetic physical-informed GRNN-XGBoost model (EPI-XGNN), based on traditional (classical) physical system and data-driven error model is proposed to more accurately describe the complex and nonlinear electromagnetic force of the stator-end winding. FE electromagnetic force, physics-informed electromagnetic force and physics-informed hybrid modeling are compared to verify the rationality of the digital physics-informed model and physics-informed hybrid modeling. In addition, experimental random noise is added to the electromagnetic force data of finite element to simulate the experimental data in the real scenarios, which verifies that the physics-informed hybrid modeling has a strong applicability in the noisy environment. The research will provide support for the fault diagnosis and prediction of the vibration dynamics of “Smart Turbine”.