Some Kinds Of Combined Forecasting Models,Properties And Applications Based On Process And Result Integration

The idea of combined forecasting is mainly to make full use of the information in each single prediction method and to integrate the prediction advantages of various single methods.This is also an effective method to improve the accuracy of prediction and reduce the risk that may be caused by a single prediction method.At present,it has been widely used in various fields of society such as financial management,environmental supervision,disease prevention and control,and policy making.The research on combination forecasting mainly focuses on the following two aspects at present.The first is,how to determine the weighting coefficients of individual prediction method under the premise of objectivity and accuracy.One of the most commonly used methods is to construct an optimization model with the smallest prediction error to find the optimal weighting coefficient,or to integrate data from a single prediction method based on information integration operators.However,these models usually only consider the information reflected by the single prediction method itself,but neglect whether the information represents overlaps between different methods.At the same time,this way may bring redundant information to the combined forecast and reduce prediction accuracy to some extent.Second,it is difficult to accurately describe some phenomenon,such as bank deposits and economic growth,in a point prediction due to the increasing complexity of the real situation.As this uncertainty and ambiguity of data increase,the prediction model ambiguity also enlarges.Therefore,more and more studies have introduced the number of intervals into a single prediction model in recent years.However,the prediction error is minimized when constructing a combined prediction model of interval numbers.The common way is to split the left and right endpoints of the interval to achieve the minimum error for the left and right endpoints,respectively,which destroys the information expressed by the interval number and violates the integrity of the interval number.Therefore,we consider the mutual influence of the information expressed by the single prediction method during the construction of the combined forecasting model,and construct a process-based integrated forecasting model.Besides,it is of great practical value and theoretical significance to discuss how to improve the accuracy of combined forecast without destroying the integrity of the interval numbers.The mainly results are as following five chapters:The first chapter briefly introduces the current status and research background of combined forecasting.It also briefly introduces the specific innovations in this dissertation.In chapter two,it is mainly introduced the concepts and arithmetic operations of interval numbers,triangular fuzzy numbers and several common Theil-inequality coefficients.In chapter three,we come up with an idea that process-based combination forecasting model,instead of optimizing model that considering how to determine the weight of each single prediction model.In the process of constructing the combination model,the output of a single prediction method is used as another method inputs.First,based on the characteristics of the data,we use the fluctuations of the data to complete the cluster objectively and divide the universe.Then,the data are synthesized based on the degree of membership between different kinds of interval numbers.Finally,two linear and non-linear models are selected and a combined prediction model is constructed based on the process and data characteristics.The example analysis shows that the proposed combination prediction model improves the accuracy of prediction to a certain extent,which shows the effectiveness of this model.In chapter four,interval number combination forecasting model,the left and right endpoints of the interval number are regarded as two different time series in general.The two sets of time series are combined respectively,which may reverse the order of left and right endpoints.This way may cause prediction error.Furthermore,combining the left and right endpoints of the interval will destroy the integrity of prediction information.To this end,the mid-point and radius of the interval are introduced to defined a new Theil inequality coefficient prediction accuracy index.At the same time,Multi-objective optimal combination forecasting model is constructed based on the midpoint and radius Theil unequal coefficients and attitude parameters are added to transform the multi-objective programming based on the coefficients of the midpoint and radius Theil inequality coefficient into a single-objective optimization model.The validity of the model is verified by an example.Finally,a sensitivity analysis is performed on the attitude parameters of the model to verify that different parameter values would affect the accuracy of the combined prediction model to a certain extentIn chapter five,the main research is to obtain the interval number prediction with higher accuracy when the predict object data is expressed by the real data.For this purpose,we propose a new concept of interval unequal coefficient prediction accuracy index and a multi-objective optimal combined forecasting model based on process and interval unequal coefficients.An attitude parameter is added to transform multi-objective programming into a single-objective optimization model.At last,the main research contents are summarized,and the research work in the future is prospected.