| Project/Area Number |
05650406
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
計測・制御工学
|
|---|
| Research Institution | Kyushu University |
|---|
Principal Investigator |
OHBAYASHI Masanao (1994) Kyushu Univ., Faculty of Engg., Research Associate, 工学部, 助手 (60213849)
村田 純一 (1993) 九州大学, 工学部, 助教授 (60190914)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
HIRASAWA Koutarou Kyushu Univ., Faculty of Engg., Professor, 工学部, 教授 (70253474)
大林 正直 九州大学, 工学部, 助手 (60213849)
|
|---|
| Project Period (FY) |
1993 – 1994
|
| Project Status |
Completed (Fiscal Year 1994)
|
| Budget Amount *help |
¥2,200,000 (Direct Cost: ¥2,200,000)
Fiscal Year 1994: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 1993: ¥1,600,000 (Direct Cost: ¥1,600,000)
|
|---|
| Keywords | Learning Petri Network / Neural Network / Self-Organization / Modeling / Modeling Error / Robust Control / Pole Assignment in a Specified Region / Loop Transfer Recovery / モデル / Loop Transfer Recovery / 学習ペトソネットワーク |
|---|
| Research Abstract |
(1) : [Self-organization of nonlinear system models]
Two types of new networks have been proposed for nonlinear system modeling. One consists of several neural sub-networks, each which is selected as appropriate route, through learning, respond to a paticular sort of input. For example, one kind of input is assigned to a certain sub-network and another is as-signed to another sub-network. The other network has combined properties of neural network and Petri network. This network is formed through learning, and realize route control automatically. Both networks have been confirmed through the identification of nonlinear system, to be more excellent than ordinary neural network.
(2) : [Modeling and control of nonlinear systems]
For nonlinear systems, the following have been proposed based on the idea that the systems can be represented by combinations of linear nominal models and neural network nonlinear models : a new neural network capable of representing both linear and nonlinear relationships, and its application to system identification and learning control ; a neural network structure designing method which eliminates redundant components ; an adaptive control scheme using the combination of a linear model and a neural network model. Their validity have been verified through simulation studies.
(3) : [Modeling, modeling errors, and control for distributed parameter systems]
For the case that the model has an error or a certain difference from the real system, a method has been proposed to design a state feedback gain that gives the system desired response by pole assignment in a specified region and minimizes a quadratic cost function simaltaneously. Another method is proposed which designs feedback and observer gains that accomplish recovery of robust stability and do the above two objects simaltaneously. Effectivness of these methods have been confirmed by numerical simulation on nonlinear crane system.
|
