Research On Modeling Of Complex Flow Behavior Based On Data-Driven

As the scale of modern power systems continues to increase,the complexity of the system continues to increase.The traditional model-driven modeling methods are increasingly challenging to meet the modeling needs.With the rapid development of computer technology and the large-scale use of sensors,massive amounts of data can be collected,stored,and analyzed,making data-driven modeling methods a research hotspot.This paper studies the basic methods and theories of data-driven modeling and conducts modeling research on compressible flow behavior and incompressible flow behavior,respectively.Data is collected from numerical simulation and experiments for modeling,and the data-driven models are compared with theoretical models.The reliability and effectiveness of the data-driven models are verified and discussed.Firstly,two data-driven modeling methods are introduced: dynamic mode decomposition and partial differential equation reconstruction methods based on sparse regression.The dynamic mode decomposition method can decompose the sampled data of the system into modals containing different system characteristics.This paper improves the dynamic mode decomposition method.The decomposed modals are reordered using energy sorting and correlation coefficient sorting methods.It can better adapt to the background of strong noise.Using dynamic mode decomposition as a data preprocessing method for partial differential equation reconstruction can reduce noise so that partial differential equation reconstruction is more accurate than not using dynamic mode decomposition.Secondly,for the Helmholtz resonance(gas compressible)systems and falling liquid film flow(liquid incompressible)systems,data is collected through three methods:numerical simulation,CFD simulation,and experimental measurement.Dynamic mode decomposition and differential equation reconstruction are composed to use for modeling research.The dynamic mode decomposition characteristics of the Helmholtz resonance system and falling liquid film system are studied separately.It is verified that the datadriven modeling method has the accuracy of frequency prediction of the Helmholtz resonance system.The differential equations reconstruction of the Helmholtz resonance equation and the integral boundary layer model equation of the moderate Reynolds number of falling liquid film are reached with the data-driven method.Finally,the problems encountered by data-driven algorithms in the modeling of complex flow systems are discussed.The existing data-driven methods for modeling complex flow systems are still only preliminary explorations and can be used as an auxiliary means of theoretical modeling approaches.The precise data-driven modeling of the flow system needs further research.