Research On Data Imputation Based On Generative Adversarial Networks Model

With the rapid development of various technologies in the information age,huge amounts of data have been generated all over the world,and a large amount of information is hidden behind such massive data.However,there are many difficulties in obtaining real data in the real world,which makes the data used for various tasks contain missing data.It is necessary to adopt appropriate data imputation methods to ensure the integrity and quality of data.At present,there are many excellent and mature methods in the field of data imputation.In recent years,deep learning and machine learning have become hot research fields,and many imputation methods based on deep learning have emerged.This paper analyzes the problems related to missing data,such as the causes of missing data,the mechanism of missing data,missing data mode,etc.inspired by the generative model,the task of missing data imputation is summarized as a problem of data generation.In this paper,the completion of data imputation is divided into two parts,they are the common type of missing data imputation and multivariate time series missing data imputation respectively.In the common type of data imputation task,the paper uses the generative adversarial network as the basic model of data imputation.On this basis,the model of the built-up countermeasure network is improved.The two variants of the recurrent neural network,respectively are the long short-term memory network and the gate recurrent unit,are used as the replacement network of the multilayer perceptron in the traditional generative adversarial network.The core component generation network and the component of the discriminant network.In the task of missing data imputation in multivariate time series data,this paper proposes two solutions to the key problem of obtaining time information in data,which are named time series distance method and sliding window method.The method of time series distance is used to improve the long short-term memory network and gate recurrent unit,so that they can capture the time information in the data.The generative adversarial network imputation model based on this method can fill the missing data of multivariate time series.In this paper,we use data sets from the real world,and evaluate the standard generative adversarial network and the build countermeasure network imputation model based on this paper.Under the average result of normalized mean square deviation,the evaluation scores of mean square error of generative adversarial network based on time series distance method are obtained on different data sets,and the results are excellent.Therefore,it is concluded that the generative adversarial network model based on time series distance method provides better performance.