| 摘要 |
PURPOSE: To easily predict a demand quantity with a high precision by learning and recognizing correlations between factor data obtained by quantization based on the degree of the influence of the actual result value on a predicted quantity and predicted quantity data generated based on the actual result value to obtain the knowledge for prediction. CONSTITUTION: In a prediction device 1, an input operation part 3, a storage part 4, and a display part 5 are connected to a control part 2 consisting of a microcomputer. The control part 2 constitutes a means which generates factor data, predicted quantity data, and the knowledge for prediction. Demand actual result data A is inputted in the input operation part 3 to generate feature data B (for each or items of the day of a week, the date, etc.). Normalized numerical data D based on the degree of influence of the actual result value is generated from data A and B. Meanwhile, the actual result value is ranked to calculate the rank likelihood degree data E of daily actual result values. Correlations between daily quantities of data D and fuzzy rank likelihood degrees are learnt and recognized, and the prediction operation and the result based on this prediction knowledge are evaluated. |
