Research On Trend Prediction Based On Time Series Of Satellite Calibration Parameters

With the development of society and technology,people are increasingly studying time series and other data.However,in reality,time series are mostly composed of several basic components.Directly using a single model to predict it,it is difficult to improve the prediction accuracy.The data studied in this paper comes from the time series of calibration parameters of the MERSI of the Fengyun 3A satellite.This time series shows an upward trend over a long period of time,and at the same time,there are periodic changes in a short period of time.It also contains noise.In view of the respective characteristics of ensemble empirical mode decomposition(EEMD)and Seasonal-Trend decomposition procedure based on Loess(STL),this paper proposes a decomposition method based on the combination of STL and the variant EEMD of empirical mode decomposition,and then based on This decomposition method uses long short-term memory network(LSTM)to model the obtained sub-sequences.At the same time,on this basis,the variational modal decomposition(VMD)is substituted for EEMD,and a specified number of sub-sequences are decomposed to obtain a new prediction model based on the mixed model.The specific research work mainly includes the following contents:1.This paper proposes a decomposition method based on the combination of STL and empirical mode decomposition of variant EEMD,that is,the data is first decomposed by STL,and after the trend term,period term,and residual term are obtained,the period term and residual are combined The items are recombined into a new time series.Then use EEMD to re-decompose this new time series to obtain multiple eigenmode functions,recombine the eigenmode functions in the order of frequency,and reconstruct new periodic terms and residual terms.Finally,the trend item,the reconstructed period item,and the reconstructed residual item are used as the LSTM data set to train the model,which improves the accuracy of time series prediction.2.Although the STL-EEMD-LSTM hybrid model has advantages in satellite sensor time series prediction,the number of decomposition modes is random.Therefore,this paper uses the variational modal decomposition VMD that can decompose the number of fixed modes to decompose the periodic term and residual term obtained by STL again,and solve the variational constraint equation that minimizes the sum of the bandwidth of the sub-sequence signal,that is,decompose the specified The number of modes is improved,and the resolution accuracy of satellite sensors is improved,and then LSTM is used for prediction,thereby obtaining a new prediction model based on the hybrid model.3.This article implements a trend prediction system based on the time series of satellite calibration parameters.Based on the theoretical knowledge and experimental analysis and results of the time series prediction model using the above-mentioned hybrid model,the process and structure of the time series decomposition and prediction are visualized.Provide help and support for the analysis and research of satellite calibration parameters.