Combination Forecasting Model Based On Multi-source Information And Multi-scale Perspectives And Its Application

With the continuous development of society and technology,prediction is becoming more and more important.The predicted object may be a complex system,the prediction risk caused by a single prediction method has tendency to increase.Combining the advantages of different single forecasting methods and the complementarity of different single forecasting methods,combination forecasting can improve the prediction accuracy and generalization ability of the model in application.In recent years,environmental problems have become increasingly prominent.How to effectively improve the ecological environment,reduce greenhouse gas emissions and improve air quality has become the focus of the world.The accurate prediction of carbon trading price and air quality index can provide important decision-making for the management to improve the environment.There are still some problems worth discussing in the forecasting research of carbon trading price and air quality index.First,most of the current studies use the influence of structured data,without considering the influence of unstructured data on the forecasting model,it will ignore the integrity of information.Second,after decomposing complex time series,most of the forecasting models used are single forecasting models rather than combined forecasting models,which may bring greater prediction risks than single forecasting model.Thirdly,with the increase of the complexity of social systems,there are few researches on the combined prediction of interval-valued time series,which is worth studying how to improve the prediction accuracy of interval number.To solve the problems above,this dissertation discusses the combined forecasting model based on multi-source information and multi-scale perspectives and its application in carbon trading price and air quality index.This research has important theoretical and practical value.The main work of this dissertation are as follows:Chapter one,mainly introduces the research significance of this dissertation,discusses the relevant research,and introduces the working ideas and innovation points of this dissertation.Chapter two,brings the definition of interval number and its related algorithms,and also introduces the general form of combined forecasting model.Chapter three is based on the existing research on combined forecasting model,the unstructured data is considered as the influencing factor,and the combined forecasting model is constructed from the perspective of multi-source information to improve the integrity of information.For the high dimension of unstructured data,the local linear embedding algorithm(LLE)is used to reduce the data dimension.Then using the least squares support vector regression(LSSVR)with parameters optimized,multi-layer perceptron(MLP)and NARX three single forecasting method,a carbon trading price combination forecast model is built based on unstructured data.Finally,an example of carbon trading price in Shenzhen is conducted,and the results show that the prediction effect is improved to a certain extent after the addition of unstructured data.Chapter four considers the increase of data complexity after adding unstructured data.The data are decomposed into sub-sequences with lower complexity and the idea of "decomposing first and then integrating",and then the sub-sequences are predicted respectively,finally the predicted results are obtained by integrating the prediction of each sub-sequences.For the reason,this chapter builds a combined forecasting model of carbon trading price based on multisource information and multi-scale perspective,and verifies that the prediction accuracy of the model has been further improved,which illustrates the rationality of the model.Chapter five extends the data types on the basis of the previous two chapters,and explores the combined prediction model of interval-valued time series.Firstly,the decomposition algorithm of interval-valued time series is improved,and the bivariate ensemble empirical mode decomposition(BEEMD)method is proposed to improve the accuracy of decomposition.Secondly three interval single forecasting methods are used to predict the decomposed data.Then an interval combination forecasting model of AQI based on the BEEMD method is built.Finally,the example of the air quality index of Hefei City is carried out.The results show that the model has high decomposition accuracy and can reduce the interval prediction error.