Title:
【発明の名称】成長機能を有するニューラルネットワークシステム
Document Type and Number:
Japanese Patent JP2986294
Kind Code:
B2
Abstract:
PURPOSE:To maintain and improve forecasting accuracy by updating the weighting coefficient of a recalling neural network by the one obtained by the learning of a learning neural network when the accuracy of a recalled result of the learning neural network is higher than that of the recalling neural network. CONSTITUTION:A weight coefficient growing means 8 started by a neural network learning means 7 reads out the measuring value (u) of time sequence data from a time sequence data memory 4 based upon the number of input neurons. A correlation coefficient between a recalled value Vc=h(Xc) based upon the weighting coefficient Wc of a recalling data memory 5 and a recalled value VDELTAT=h (XDELTAT) based upon a weighting coefficient WDELTAT is found out and accuracy evaluation is executed by the correlation coefficient. When the forecasting accuracy of the recalled value of the weighting coefficient WDELTAT obtained in the past of DELTAT is higher than that of a weight coefficient obtained at the preceding time, the means 8 can change the weighting coefficient through an on-line when necessary.
More Like This:
Inventors:
KOIDE MINORU
INOE HARUKI
YAHIRO MASAKAZU
ABE NOBORU
INOE HARUKI
YAHIRO MASAKAZU
ABE NOBORU
Application Number:
JP32092692A
Publication Date:
December 06, 1999
Filing Date:
November 30, 1992
Export Citation:
Assignee:
HITACHI SEISAKUSHO KK
HITACHI ENJINIARINGU KK
HITACHI ENJINIARINGU KK
International Classes:
G06F15/18; G06G7/60; G06N3/08; G06N99/00; (IPC1-7): G06F15/18; G06G7/60
Domestic Patent References:
| JP4199259A | ||||
| JP4205163A |
Attorney, Agent or Firm:
Kazuko Tomita