Research On Virtual Sample Generation Technology Based On Quadrat And Quantile Regression

In the context of the increasing popularity of informatization,a large amount of data is generated and preserved,and it has always been a research hotspot to dig and analyze the underlying laws.However,due to the high cost of data acquisition,low probability of occurrence,and small data fluctuations,in some fields,the number of representative samples with research value is small,and the problem of small samples is prominent.This is also a very critical issue for the use of data-driven modeling.The quality and quantity of training samples play a key role in the generalization learning ability and model accuracy of the model.The virtual sample generation method,as a technology to effectively solve the problem of small samples,has received widespread attention.This paper proposes a virtual sample generation method based on quadrat and quantile regression methodology(QQRMVSG).The main idea is to divide the input space by quadrat squares based on Dominance Analysis,and generate virtual samples in the input space to make sure that there is a virtual input in each quadrat square.Then the virtual outputs corresponding to the virtual inputs are predicted by Gaussian process regression.Finally,the correlation between the independent variables and the dependent variable is analyzed by quantile regression,and the virtual samples that do not conform to this relationship are eliminated.In order to increase the number of generated virtual samples,the Local Outlier Factor algorithm is applied to QQRMVSG.The least outlier sample points in the input space are used as the initial points for generating virtual inputs process and checked points for screening process.Add the random mechanism to make the virtual samples generated differently each time,and finally increase the number of virtual samples.Finally,the standard function data set and the high-density polyethylene(HDPE)industrial case are used to verify the rationality and effectiveness of QQRMVSG.