Research And Application Of Time Series Combined Forecasting Models

With the rapid development of science,technology,and social economy,the data and information of all industries are exploring,meanwhile,the emerging of the Internet intelligence technology marks the era of big data.How to seize the opportunities,specifically to extract valuable information from massive data so as to make an accurate prediction of the future is a meaningful task.In reality,a large amount of data and information are time series data.Conventional time series prediction means to make the quantitative estimation of the future trend based on historical values and current values of time series by utilizing statistical models.Recently,domestic and foreign scholars have made many valuable explorations and researches in the field of time series prediction.However,due to the universality of time series data and the complexity and variability of such problems,till now there is no one research that is powerful enough to address the prediction problems of various time series data.The existing time series forecasting models can be roughly divided into two categories,namely the traditional single model and the combined prediction model.For the combined forecasting model,according to the logical attributes,it can be further divided into a deterministic combined forecasting model and a non-deterministic combined forecasting model.However,the current research on various combinations of predictive models has its own shortcomings.Specifically,there are still some issues that need further research:firstly,deterministic combined models,the primary purpose of which is to provide effective deterministic predictions for time series data in reality.Because of the various characteristics of data generated by different fields and industries,the requirements of the prediction efficiency of the model are also different.Therefore,it is necessary to establish different prediction models to adopt different data;secondly,the non-deterministic combined model.Nowadays the modifications of the fuzzy time series model mainly focus on the fuzzifying process,the establishment process of the fuzzy relation matrix and the process of defuzzifying,while there are fewer researches to pay attention to the weight information in the fuzzy logistic relationship,however,it can affect the results of the prediction;thirdly,rationality of model effect.For different types of prediction models,linear or nonlinear,single or combined,deterministic or non-deterministic prediction models,it is necessary to study the rationality of the model effect.It is found that the predictability of time series data has a tendency relationship with the source structure of its multifractal.Therefore,based on the fractal theory,the predictability of data can be inferred from the multifractal parameter characteristics of the data,then the rationality of the prediction model can be tested.Based on such status and existing problems,this paper conducts the research and application of time series forecasting models in multiple perspectives.This research focuses on the combined forecasting theory and fractal parameter characteristics of time series data,by using the information fusion theory,intelligent combined method,fuzzy time series model,multifractal method and so on.For a different type of data with various characteristics,this paper establishes corresponding deterministic or non-deterministic combined forecasting models,which will be applied in different practical problems background,after that the rationality of the combined forecasting model will be verified by the fractal parameter characteristics of data,which is obtained from fractal theory.The whole content of this paper can be divided into six parts:The first chapter will introduce the background and the basis of the topic,the research content,and significance,as well as the main innovations and deficiencies.The second chapter will summarize the researches done by domestic and foreign scholars.By making comparisons and comprehensive reviews,the pertinence and rationality of this research are clarified.The third chapter is the establishment and application of the deterministic combined prediction model based on information fusion.The empirical process shows that the deterministic combined prediction model performs well and rational in the context of air pollution practical problems;for a pure time series data without external information,the fourth chapter establishes a deterministic combined prediction model based on intelligent methods,the empirical research carried out in the background of wind power output power prediction shows that the combined forecasting model also has a good and rational model effect;the fifth chapter is the establishment and application of the non-deterministic combined forecasting model based on fuzzy theory.The empirical process shows that the non-deterministic combined model works well and rationally in different fuzzy contexts;the sixth chapter is the conclusion of this paper,as well as the prospects for future research Specifically,the conclusions of this paper mainly include the following three aspects:First,the deterministic combined prediction model based on information fusion has a good and rational model effect.The basic idea of the combined prediction model is to maximize the utilization and fusion of the information around the target,and to slim down the training set through the hierarchical strategy,so as to achieve effective training and prediction of the combined prediction model.The combined prediction model shows a good model effect in the empirical process of the air pollution problem,and the rationality of the combined prediction model is verified by analyzing the fractal parameter characteristics of pollutant data.This research and findings can act as an early warning tool for management,which can provide effective prediction and theoretical reference and mathematical support for the implementation of relevant policy decisions.It has strong practical meaning for China's air pollution prevention work and countermeasure research.Secondly,for the single time series data without considering external information,a deterministic combined forecasting model based on intelligent method is constructed,and the combined forecasting model has a good and rational model effect.In the empirical process of wind power prediction,the deterministic combined prediction model based on intelligent method shows a good model prediction effect,and by using fractal theory to analyze the fractal parameters of data,and based on the research findings that the fractal parameters of data have a relationship with predictability,the rationality of the proposed combined prediction model is verified.This research can not only benefit the effective management of the power grid but also helps to alleviate the unstable factors in the growing wind energy field,to optimize the bidding strategy in the power market,which has a certain meaning for the safe and efficient development of the wind power industry.Finally,the non-deterministic combined forecasting model based on fuzzy theory has a good and rational model performance in the context of fuzzy data.The combined prediction model effectively combines the harmony search algorithm,the RIM weighting method,the ordered weighted aggregation operator and considers the overall logical weight information in the fuzzy logical relationship group.The empirical analysis process based on data with different characteristics shows that the proposed non-deterministic combined prediction model has high model stability and good prediction effect.Furthermore,by using fractal theory to analyze the fractal parameters of data,and based on the research findings that the fractal parameters of data have a relationship with predictability,the rationality of the proposed combined prediction model is verified.Based on different empirical evaluation and analysis,the proposed non-deterministic combined model can be applied in many fields,not only in the financial field but also in the industrial field.The main innovations of this paper are as follows:(1)based on the data of China's urban air pollution and the output power data of wind farms in western China,a deterministic combined forecasting model based on information fusion and a deterministic combined forecasting model based on intelligent method are proposed and established for the differences of data in different practical problems.,and they are proved to have good model performance through empirical analysis in the background of practical problems;(2)this paper proposes and establishes a non-deterministic combined forecasting model based on fuzzy theory,and it has good effect in two different empirical processes,by using fractal theory to analyze the fractal parameters of data,and based on the research findings that the fractal parameters of data have relationship with predictability,the rationality of the proposed combined prediction model is verified,which makes up for the gap of the logical weight information in the fuzzy logic relationship group in the existing research;(3)different from the previous model establishment and application researches that are studied on the actual problem background directly,this paper studies the construction of time series combined forecasting model and the empirical process,meanwhile the analysis of the characteristics of data fractal parameters is carried out in the background of different practical problems,which is utilized to verify the rationality of forecasting models,this can also provide the necessary reference for model construction and theoretical research of real problems.The main deficiencies of this paper are as follows:(1)the combined model established for different practical problem backgrounds may use many single methods and auxiliary tools.In fact,there may be better alternative methods and tools,but due to space and time limitations,this paper did not do further research;(2)the data used in the study are collected from specific practical problems,and if there are more data can be used in the study,the research process may be more abundant and the conclusions may be stronger.