PURPOSE: To optimize the method and system configuration for learning and forecasting a neural net by providing the method and system with an input unit setting means and neural net determination learning means in conformity with the degree of error of a learning means and learning result evaluating means for evaluating learning condition.
CONSTITUTION: The present neural net learning and forecasting method and system includes an evaluating learning means 1000 for evaluating the learning condition, a learning result evaluating means 2000, an input unit setting means 3000 to be done in conformity with the degree of error, and a neural net determination learning means 4000. Further, from the evaluating learning means 1000, a learning result 1500 is output to the learning result evaluating means 2000, and the learning result evaluating means 2000 outputs the result of evaluation 2500 to the input unit setting means 3000 or to the neural net determination learning means 4000. Accordingly, a neural net necessary to obtain a desired output can be automatically set up, and by using the neural net for demand forecasting, the forecasting precision can be enhanced.
MATSUMOTO HIROSHI
YAMADA AKIHIKO