Title:
CONTROL PARAMETER GENERATOR USING NEURAL NETWORK
Document Type and Number:
Japanese Patent JP3249157
Kind Code:
B2
Abstract:
PURPOSE: To make it possible to precisely identify the dynamic characteristic of a control object in a state that a disturbance is added and to generate a control parameter value satisfying the control specification of a real plant.
CONSTITUTION: This control parameter generator using a neural network is provided with a neural network 11 capable of outputting the control parameter of a controller by calculating an input time series signal and learning by the correction of a synapse load to be a calculation factor, a normalized signal generation element 12 generating a normalized signal showing the response of a feedback control system, an input signal changeover element 13 inputting the time series signal of the feedback control system when the the characteristic of a control object is learned or inputting the same kind of normalized signal as when the learning is performed when the control parameter value to be the control object is determined and a learning mechanism 14 correcting the synapse load based on the output error of the neural network 11 for the control parameter value of a controller when the learning is performed.
Inventors:
Ryuichi Kuwata
Application Number:
JP30191191A
Publication Date:
January 21, 2002
Filing Date:
November 18, 1991
Export Citation:
Assignee:
Toshiba Corporation
International Classes:
G06F15/18; G05B13/02; G06G7/60; G06N3/00; G06N99/00; (IPC1-7): G06N3/00; G05B13/02
Domestic Patent References:
| JP3233702A | ||||
| JP511805A |
Attorney, Agent or Firm:
Takehiko Suzue
Previous Patent: NEGATIVE-PHASE PROTECTION POWER CONTROL
Next Patent: METHOD FOR VACUUM HEAT TREATMENT OF ACTIVE METAL
Next Patent: METHOD FOR VACUUM HEAT TREATMENT OF ACTIVE METAL