Research and Implementation of Missing Value Interpolation for Time Series Data
The traditional interpolation algorithm for missing data in time series has low stability,and the error of the interpolated data is large,which can not guarantee the authenticity of the interpolated data set.Therefore,this pa-per proposes an improved interpolation algorithm based on GAIN network data,and constructs a GAIN-W interpolation model for missing data in time series by fusing Wasserstein discriminant.The first module improves the operability of the data through data dimension reduction and standardization processing,and extracts the double-chan-nel two-dimensional characteristics of the time series data based on an RW sliding window method.The second mod-ule uses a two-dimensional convolution layer to replace the fully connected layer of the generator,so as to improve the accuracy of uneven interval feature processing,and optimize the loss function of the generator by increasing the hyper-parameter λ.The third module solves the problem of gradient explosion of the discriminator based on three groups of fully connected layers,and uses Wasserstein discriminant to learn data distribution and complete data interpolation.Simulation results show that the average RMSE error of GAIN-W is reduced by 47.00％ and the average MAX error is reduced by 24.00％ compared with six baseline algorithms in HEPC and BJPM2.5 open data sets,which indicates that the GAIN-W algorithm has high accuracy and stability.To sum up,the GAIN-W trajectory missing data interpolation algorithm solves the problem of uneven timing and reduces the error of interpolation data,which has im-portant simulation research value.
DiscriminantTime series dataMissing data interpolation
NETL
NSTL
万方数据
