Research On Forecasting Model Based On Missing Data With Mixed Frequency Extreme Learning Machine

In this paper,we study the forecasting model of the mixed frequency extreme learning machine based on missing data,which refers to the mixed frequency model with different time and frequency variables when the data is missing.The purpose is to deal with the problem of management data,when missing data occurs and there is mixed frequency data.In the context of today's big data era,the rapid development of mobile Internet and Internet of Things,the source and access of data and the rapid growth of growth,often accompanied by incomplete sample data obtained during data collection,obtained samples.The problem of inconsistent acquisition frequency between data.The occurrence of these problems has received much attention.Summarizing the research of mixed frequency problems faces the following main problems:(1)In the process of data collection,the lack of data often occurs,which makes it impossible to effectively predict the research objects.(2)In real life,there are often cases of independent and time-variant data sampling models with different time statistics.(3)In the research of the existing mixed frequency model,it is found that for the study of the mixed-predictive model of the one-ary independent variable,due to the different lengths of the two ends of the data,with the increase of the data of the unary independent variables,the one-dimensional independent variable mixed frequency forecasting model exhibits nonlinear characteristics,resulting in inaccuracies in the forecasting results.(4)Similarly,when the existing multivariate mixed frequency forecasting model is studied,as the independent variable changes and increases,there may be inconsistencies in the mixed frequency length and other factors between the multivariate independent variables and the independent variables.There are also more complex nonlinear and uncertain relationships between independent and dependent variables,leading to more and more complex problems in processing data.Therefore,for the first time,this paper attempts to integrate gray absolute correlation theory,mixed frequency data forecasting theory and extreme learning machine theory and deal with the lack of data and the mixed frequency problem of data in the process of energy demand forecasting management.Explain the main research work of this paper from three aspects:First,establish a one-dimensional mixed frequency extreme learning machine regression forecasting model for complete data.Among the existing mixed frequency forecasting models,the most commonly used one-dimensional mixed frequency MIDAS model.However,during the research,it is found that the construction of the existing mixed frequency MIDAS model is always based on the discussion of time series regression forecasting and the main application is based on the research of financial market and macroeconomics,aiming at the existing one-dimensional mixed frequency MIDAS model.In other words,as the dimension of the independent variable increases,the nonlinear characteristics become more and more obvious and the forecasting error will become larger and larger.At the same time,for the one-dimensional independent variable mixed frequency MIDAS forecasting model,the mathematical derivation changes show that the independent variables are always developed in one-dimensional form.This form ignores the one-ary independent variable and the independent variables may exist between them.The associated information has certain limitations.Aiming at the above deficiencies,a model of one-dimensional mixed frequency extreme learning machine is proposed and the original one-dimensional mixed frequency MIDAS data forecasting model is extended.The new one-ary mixed frequency model is applied to China's energy demand analysis and the comparison model forecasting results are analyzed to prove the proposed new model rationality.Second,a regression forecasting model based on missing data for one-dimensional mixed frequency extreme learning machine is established.In objective reality,missing data often occurs,and it is necessary to study the mixed frequency MIDAS model with missing data.The occurrence of missing data is rarely studied in the mixed forecasting model.In order to solve the situation of mixed frequency sampling missing data in objective reality,this paper establishes a one-dimensional mixed frequency extreme learning machine forecasting model based on missing data.First,when conducting a predictive study,the effect of an independent dependent variable that may have missing data on the predicted dependent variable outcome is considered.Secondly,the unary mixed frequency MIDAS forecasting model with missing data cannot directly predict the result itself and there is a nonlinear relationship between the variable data.The grey absolute correlation analysis model is an effective model for dealing with missing problems.By judging the similarity between the two groups of data sequences,the corresponding associations are identified and the missing data is filled.In this paper,the grey absolute correlation analysis model and the one-dimensional mixed frequency extreme learning machine forecasting model are effectively integrated and the new model is applied to the Chinese energy demand forecasting problem.The predicted unary independent variable contains missing mixed frequency data and the forecasting result is compared with the comparison model forecasting result.Prove the validity of the proposed model.Thirdly,a multivariate mixed frequency extreme learning machine regression forecasting model based on missing data is established.There are often more than one independent variable in the process of predictive analysis and the independent variable and the independent variable,the independent variable and the dependent variable always interact with each other.Through the study of the mixed frequency model,it is found that with the independent variable research dimension and multivariate,The increase in the amount of data information of variables has great limitations with the existing multivariate mixed frequency MIDAS forecasting model and is often accompanied by the occurrence of missing data.Therefore,the establishment of a multivariate mixed frequency forecasting model with missing data is the focus of research.Based on the multivariate mixed frequency MIDAS forecasting model,a multivariate mixed frequency extreme learning machine forecasting model based on missing data is constructed.The problems to be solved are as follows:(1)Mixed frequency between multiple independent variables and independent variables,independent variables and dependent variables.After the data dimension is increased,there will be more complex nonlinear feature relationships and the estimated parameters will increase significantly.(2)Multiple missing values may occur after multiple independent variables and the mixed frequency data dimension between the independent variable and the independent variable and the dependent variable increase.In order to solve the above shortcomings,a multivariate mixed frequency extreme learning machine forecasting model based on missing data is proposed.The new model is applied to China's energy demand forecasting problem.By using the multivariate mixed frequency data with missing values,it is compared with the comparison model forecasting results.The validity of the proposed model.