Title:
DEVICE AND METHOD FOR NEURAL NETWORK TYPE NON-LINEAR FLOW CONTROL
Document Type and Number:
Japanese Patent JP3144658
Kind Code:
B2
Abstract:
PURPOSE: To divide and learn the neural network so that it can be applied to non-linear flow control of a large scale network by executing a processing adapted to topology of a network by a flow calculating means, and also, providing a communication means between the adjacent flow calculating means.
CONSTITUTION: A flow calculating device 71 calculates a relation of a flow and density in plural dividing points for dividing between nodal points of a network. A communication means 72 is a communication means provided through local coupling between the adjacent flow calculating devices 71. A learning device 73 determines a relation of a flow and density and a value of local coupling so as to be adaptive to the flow observed actually, so that matching of the flow observed actually, and the flow calculated by the flow calculating device 71 is taken. A controller 74 connects and disconnects the local coupling between units so that a value of a target function of control takes an extreme value. In such a way, a large scale network can be applied to non-linear flow control.
Inventors:
Tadashi Uchiyama
Noboru Sonehara
Yukio Tokunaga
Noboru Sonehara
Yukio Tokunaga
Application Number:
JP27999792A
Publication Date:
March 12, 2001
Filing Date:
October 19, 1992
Export Citation:
Assignee:
Nippon Telegraph and Telephone Corporation
International Classes:
G05B13/02; G06F9/44; G06F15/18; G06G7/60; G06N99/00; (IPC1-7): G05B13/02; G06F9/44; G06F15/18; G06G7/60
Attorney, Agent or Firm:
Hidekazu Miyoshi (1 outside)