The Classification of High Dimsnsion Flew Field Based on Manifold Learning
|School||Harbin Institute of Technology|
|Course||Power Machinery and Engineering|
|Keywords||hypersonic inlet pattern recognition manifold learning|
The main tasks of unstart protection control for the hypersonic inlet include two aspects:one is "early warning before unstart":while inlet in its normal working hours, through effective early warning methods and means to prevent the occurrence of the inlet did not start;the other is "restart control":once the inlet is not a starter, to take effective measures to re-start the inlet.Analysis of these two control tasks tell that, in order to achieve early warning before unstart and restart control, the primary mission is analysis of the inlet working model for testing.This is starting to achieve the protection control of the unstart inlet.Study found that the inlet different mode of state detection can be summed up as inlet starter / non-starter multi-mode classification issues.If we could establish the inlet between the different models and criteria for the classification model, we can work on the inlet of the current model’s detection and judgement, This is also the prerequisite for the protection control of the inlet unstart.In this paper, we research in the hypersonic inlet starter / non-starter multi-mode classification problems, first of all establish the flow numerical simulation platform.We Detailed on the Hypersonic Inlet numerical simulation, used the method to simulate a typical inlet and compared verify with the experimental data,at last we work on a numerical simulation Hypersonic inlet.Conditions in a recorded by common measurement points the flow of data ,which exist dimension too high and calculated complexity higher difficulty, is not conducive to real-time data classification discrimination.This paper introduced a dimension reduce method based on non-linear thinking.We described the manifold learning methods first,and used this method with the high-dimensional data and research in its visual analysis.By adjusting the coefficient, compare the difference of the visual effect ,we descript the manifold learning visualization methods in dealing with this subject on the feasibility and effectiveness.Finally, this paper study in the classification of the high dimesion flow data based on the manifold learning.We proposes a definition of distance with this physical objects, and reserch the effect of the classification based on manifold learning;We also proposed a weighted embedded in the manifold learning for the classification,and analysised the complexity of calculating the cost and time.