1995 Fiscal Year Final Research Report Summary
Modelling and Control of Nonlinear Distributed Parameter Systems
| Project/Area Number |
06650478
|
|---|
| 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 | Keio University |
|---|
Principal Investigator |
SHIMAZU Kiyotaka Keio University, Faculty of Science and Technology, Professor, 理工学部, 教授 (50051545)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
HAMADA Nozumu Keio University, Faculty of Science and Technology, Professor, 理工学部, 教授 (80051902)
KUNIMATSU Noboru Keio University, Faculty of Science and Technology, Professor, 理工学部, 教授 (70051662)
|
|---|
| Project Period (FY) |
1994 – 1995
|
| Keywords | semigroup, / inertial manifold, / nonlinear optimal control, / multi-layred neural network, / steepest descent learing algorithm, / global opitimization, / piecewise linear model, / fuzzy control |
|---|
| Research Abstract |
We investigate the modelling and design of nonlinear (semilinear) distributed control systems via the three methods of the following ;
(1) semigroup model, (2) neural network and arry variables, (3) Fuzzy model
(1) Semigroup model
We showed the way of how to construct the finite dimensional linear compensator for the linear semigroup model under nonlinear perturbation and showed its robustness to the perturbation. Also, we made clear that the inertial manifold exists for the over-all control system (dynamic comoensator plus semilinear system) under some additional conditions on spectrum distrubution for linear princiak part and that the manifold can be used to make the refined dimensional stabilizer.
(2) Array variable
We showed that calcuation of optimization could be effectively excuted using array variables when we implemented a steeoest descent learing algorithm. We proposed the finite element network as a neural network and investigted methods for system modeling by using it.
Generally, optimal cantrol problems for nonlinear systems necessitate minimizing a performance index defeined with a multimodal function. So we developed a quasi-steepest descent method using chaos to achieve global opimization by multi-layred neural networks.
(3) Fyzzy model
For the Takagi-Sugeno model of given nonlinear plant, piece-wise linear control system in the state-space domain is developed. In order to cope with modelling error, robust quadratically stabilizing method and fuzzy inference are applied.
|
