Some Combination Forecasting Models And Their Applications Based On Fuzzy Information Aggregation Operators

Combination forecasting can improve accuracy and spread risk of forecasting effectively, which is obtained widely application in the fields of social-economic, eco-environment and management, etc.Currently, there are two problems in the study of combination forecasting model primarily. For one thing, the key step of combination forecasting is to determine weighting coefficient of every single forecasting methods, and aggregate results of single methods into a combination forecasting value. In existing combination forecasting models, the process of aggregation is usually realized by information aggregation operators, which has made remarkable achievements. However, those models only consider the information itself obtained by single forecasting methods, and ignore interactive relationship among them. For another, in previous combination forecasting models, taking the form as determined real numbers, weighting coefficients and forecasted information cannot reflect performance of each single forecasting method comprehensively; and with the increase of uncertainty and complexity in social-economic system, it is necessary to describe objectives with fuzzy data in different levels.Therefore, it is worthy to study the problems existed in combination forecasting models and research how to consider the interactions of all kinds of single forecasting methods and construct the fuzzy combination forecasting models are theoretically and practically. This paper focuses on the problems mentioned above, and discusses four respects of combination forecasting models as follows:(1) The continuous interval combination forecasting model based on uncertain weighted power averaging (UWPA) operator is constructed applying power operator and continuous ordered weighted (COWA) operator. Meanwhile, concepts of non-inferior and superior combination forecasting are proposed, and non-inferior property of new model is proved. Moreover, the illustrated example demonstrates validity of this method, and sensitivity analysis of parameters is also accomplished.(2) By generalizing interval combination forecasting model, a fuzzy optimization combination forecasting method is proposed, where forecasted information takes the form of triangular fuzzy numbers. At the same time, considering interactive relationship among forecasted information, the combination forecasting model based on triangular fuzzy weighted power (TFWP) operator is proposed. Additionally, the property of model is discussed, and its strong application background is demonstrated by an example.(3) A novel combination forecasting model is developed through generalizing weighting coefficients from real numbers to triangular fuzzy numbers. Meanwhile, considering that forecasting accuracy varies with time points and methods, the induced ordered weighted averaging (IOWA) operator is applied to propose the fuzzy continuous interval combination forecasting model based on induced continuous ordered fuzzy weighted averaging (ICOFWA) operator. The method avoids employing weighting coefficients by determined real numbers, improves limitation of previous methods, and its application is demonstrated by an example.(4) Through generalizing both weighting coefficients and forecasted information to triangular fuzzy numbers, a new fuzzy optimization combination forecasting model and a new definition of forecasting measure are proposed. Furthermore, based on induced ordered triangular fuzzy weighted averaging (IOTFWA) operator, the multi-object fuzzy optimization combination forecasting model is developed, which is then transformed to single object problem to calculate. Additionally, sensitivity analysis of parameters is also accomplished, and the example of power load forecasting demonstrates feasibility of this new method.