| 摘要 |
PURPOSE:To obtain a highly precise prediction value by extracting data where similarity can be recognized from time-sequantial data and reconstructing attracters for every generated similar data group. CONSTITUTION:Time-sequential data and external cause data are stored in a data storage part 1, and a similar data generation part 2 divides time- sequential data into the groups different in attracter characteristics from each other. A division attracter reconstitution part 3 reconstructs the divided attracters for every divided similar data group. A fuzzy rule generation part 4 finds the range of the similar data groups according to what range of values plural external causes are in. A realized degree calculation part 5 specifies real data of an event to be predicted, and specifies external cause data at the same time. The values of the realized degrees of rules are obtained by the individual rules, and a predicted value generation part 6 applies real data to the respective divided attracters and calculates the divided prediction values in the respective divided attracters. |
