Research on surrogate models and optimization algorithms of compressor characteristic based on digital twins

Abstract

For the digitization of turbocharger, the prediction of compressor working state is essential. How to build a model with accurate prediction and less time-consuming is the premise of studying the digitization of turbochargers. As the relationship between compressor parameters is obtained through experiments, it cannot be expressed by simple functional equations, so the surrogate model is often used for fitting the curve. Five surrogate models, the Kriging model, Response Surface Methodology, Artificial Neural Networks, Radial Basis Function, and Support vector machines, were used to fit and regression compressor characteristic curves. And four optimization algorithms, Particle Swarm Optimization, Genetic Algorithm, Gray Wolf algorithm, and Firefly Algorithm, were used to optimize the model. A method to construct a hybrid surrogate model is proposed. The results show that the influencing factors of the modeling pressure ratio and efficiency at all speed groups were confirmed; Different optimization algorithms have different optimization degrees for the five surrogate models; The prediction accuracy of the hybrid surrogate model is better than the optimized model and the single model. The constructed model can be applied in the digital twins system to predict the working state of the compressor in time to achieve the purpose of rapid response.