Research And Modeling Application Of VSG Method Based On Monte Carlo And Kernel Density Estimation

With the gradual expand of Internet technology,we have ushered in the epoch of big data.Although big data has infiltrate into nooks and crannies of modern life,there are still small sample problems in some areas.The small sample problem is mainly reflected in the difficulty of obtaining sample data,lack of data and uneven data distribution,which bring matter about poor model performance and poor generalization ability.Therefore,it is very necessary to put forward a method to solve the small sample problem and establish a stable and reliable model.The current solution to the problem of small samples is mainly virtual sample generation technology.In order to generate more reasonable virtual samples,this article put forward a VSG method based on output sample KDE and Monte Carlo to alleviate the problem that data with uneven distribution brought by small samples.The proposed method uses KDE to fit the output distribution,Monte Carlo gains the virtual output and then acquires the virtual sample input through the BA algorithm and the random weight neural network based on regularization.After obtaining the overall virtual sample.The residual analysis removes the bad virtual samples and obtains the final virtual sample set.The method is validated by two industrial data sets:MLCC and PTA.The trial bear out the VSG by the advance theory can effectively fill the gaps between samples and better alleviate the small sample problem in the actual industrial production process.In addition,the article also put forward a Bagging model based on multiple neural networks.Using extreme learning machines and random weight neural networks based on regularization and using PCA for the output data of the hidden-layers in the network to further optimizes the model.The theory is validated by the MLCC and PTA.The trial bear out the model’s performance is better.