Research On Superiority Interval Type Combination Forecasting Model Based On Correlation

Since the data samples are heterogeneous,the research object of prediction method therefore should be differentiated.In some cases,the prediction methods have been overall well matched,however,at some particular moments,the prediction errors are still large.In this paper,we attempt to combine a group of different single-item forecasting methods,then adding a weighting factor to establish a combined predictive model.It is expected that the model could incorporate effective information from individual predictive methods,adopt their merits,and avoid their shortcomings and therefore improving the accuracy of prediction.The pivot point in forming a combined forecasting model is that how to construct an appropriate target criterion and information integration operator which enables comprehensive measure of assessing and calculating the weight of individual forecasting methods in the combinational prediction,and thereby achieving a more accurate estimation.Currently,combined forecasting methods are mostly applied to real number sample data.However,issues such as the uncertainty of data or the subjectively fuzzy thinking of the observers may thwart people from describing the research object with relatively accurate data.To address these issues,this paper attempts to replace the real number sequence with the interval number sequence which makes the combination forecasting method more meaningful.In order to explore the research object of interval data type,this paper constructs six interval number combination forecasting models,and the accuracy is examined by both theoretical and empirical evidence.The key points of this paper are summarized as follow:(1)By employing the gray relational grade as the target criterion for the midpoint and the radius of interval number,this paper target to construct the multi-objective model of interval combination forecasting based on gray relational grade.Additionally,to simplified the model,coefficients of preference will be introduced into the model.Furthermore,this paper includes the combination forecasting model of weight coefficient,the combination of variable weight coefficient of GIOWA operator,and the validity theory of weighting coefficient model: theoretical study is made on the combination of non-inferiority interval numbers,whether there are redundant single method in the combination of the judge theorem,redundant information to determine the theorem.Moreover,this study selects an appropriate interval value of social security level as a sample,and it estimates the effectiveness of the two models by an example(2)This paper constructs a new indicator of interval number as a target criterion and researches the validity theory of weighting coefficient model by establishing two separate models,namely the optimal combination forecasting model based on interval number correlation and another combination forecasting model based on interval number correlation and GIOWA Operator Variable Weights,and the validity theory of weighting coefficient model: theoretical study is made on the combination of non-inferiority interval numbers,whether there are redundant single method in the combination of the judge theorem,redundant information to determine the theorem.Meanwhile,an appropriate interval value of a provincial social security level is chosen as a sample to practically test the validity of these two types of models.(3)After converting the interval number into binary numbers and constructing a new binary relevance index as a target criterion,it researches the validity theory of the weighting coefficient model by establishing two separate models,namely the optimal combination forecasting model based on the relevant number correlation and another combination forecasting model based on the relevant number correlation and GIOWA Operator Variable Weights,and the validity theory of weighting coefficient model: theoretical study is made on the combination of non-inferiority interval numbers,whether there are redundant single method in the combination of the judge theorem,redundant information to determine the theorem.Meanwhile,an appropriate interval value of a provincial social security level is chosen as a sample to practically test the validity of these two types of models.