Research And Development On The Model-Selecting-Rule-Based Combination Forecasting Support System

With the further development of market economy in our country, demands forecasting becomes an important activity for an enterprise to be exist and develop. Demands forecasting is not only the base of enterprises decision-making, but also the important gist for planning resources. So that forecasting is the precondition for an enterprise to address its strategy and tactic. Since forecasting could help enterprises meeting with demands and improving their competitive power, it is necessary that we study demands forecasting theories and combine them with information technologies so as to improve the level of demands forecasting so as to drive a quick and reasonable decision in every tache of the activities in the enterprises and make their competitive power increased.The results of forecasting theories research indicate that combination forecasting based on some rules is much effective in enhancing forecasting precision and decreasing forecasting risks. Therefore, it would made significant sence for an enterprise to address her strategy and to improve the level of decision-making if a demands forecasting system supported by rule-based combination forecasting theories were set up. However, there almost hasn't any combination forecasting support system exists although a lot of forecasting support systems or forecasting expert systems have been under research and some commercial forecasting software has been provided.In order to imporve the level of research on demands forecasting theories and computerial application of the forecasting support system, a development theory of a model-selecting-rule-based combination forecasting support system is proposed in this article. The idea not only brings various combination forecast models into the sytem to enhance the system's ability in decision support with the help of the advantage of combination forecasts, but also constructs a rule base with the gist of forecasting experiences to help the users selecting a combination forecast model with more effectiveness. First of all, the results of the research on forecasting theories and systems are summarized and the meaning of the study in this article and what existent studies help in this article are pointed out. Secondly, the theories of forecast are summarized and the advantage of combination forecast comparing with individual forecast is analyzed. Thirdly, the basic theory of combination forecast based on model-selecting rules is brought forward. The general formation of the combination-model-selecting rules and the top work flow of combination forecast instructed by the model-selecting rules are labored. In order to ensure that the rules are adaptive in differnet forecasting circumstances, the adaptive mechanism of the model-selecting rules is discussed with the help of statistic mathods. Fourthly, the development theories of the model-selecting-rule-based combination forecasting support system are put forward using the basic theories proposed above. The work flow and architecture of the prototype system are analyzed and the constructions of the model base and the rule base are discussed in detail. Fifthly, the development theories are applied on the demands forecasting support system. The work flow and data relationships are analyzed and the function, architecture, three bases and user interfaces are designed. Then the system was deployed and run under simulation environment and the effects are analyzed. At last, the author's work is summerized and the directions for further research are pointed out.The ideas of the formation of the model-selecting rules, the construction of the rule base, the adaptive mechanism of the rules and the development of the model-selecting-rule-based combination forecasting system not only integrate the quantitive forecast models with the qualitative forecast experience but also combine the forecast theories with the information system theories. They are effective in improving forecast speed, veracity and practicability and decreasing forecast risks. These theories have no doubt proposed new ideas to the research of forecasting theories and systems.