• Transactions of the Korean Society of Mechanical Engineers
• /
• v.18 no.6
• /
• pp.1542-1550
• /
• 1994

For a class of singularly perturbed uncertain system, an output feedback control law is designed. The controller structure is designed based on the uncertain reduced-order system, and the controller parameters are determined by information on the reduced-order and full-order systems. It has been shown that the reduces-order system with the designed controller possesses a stability property(specifically, a global uniform attractivity). Furthermore, the stability property of this control scheme is robust with respect to singular perturbation ; i.e., the full-order system, subject to the same controller, possesses the global uniform attractivity, provided the singular perturbation parameter $\mu<\mu^{*}$, where a threshold value $\mu^{*}$ can be computed from the information available on the full-order system.

• 제어로봇시스템학회:학술대회논문집
• /
• 1995.10a
• /
• pp.1-4
• /
• 1995

We propose a new design method for a generalized predictive control (GPC) system based on the parametrization of two-degree-of freedom control systems. The objective is to design the GPC system which guarantees the stability of the control system for a perturbed plant. The design procedure of our proposed method consists of three steps. First, we design a basic controller for a nominal plant using the LQG method and parametrize a whole control system. Next, we identify the deviation between the perturbed plant and the nominal one using a closed-loop identification method and design a free parameter of parametrization to stabilize the closed-loop system. Finally, we design a feedforward controller so as to incorporate GPC technique into our controller structure. A numerical example is presented to show the effectiveness of our proposed method.

• Journal of applied mathematics & informatics
• /
• v.22 no.1_2
• /
• pp.49-65
• /
• 2006

A robust numerical method for a singularly perturbed second-order ordinary differential equation having two parameters with a discontinuous source term is presented in this article. Theoretical bounds are derived for the derivatives of the solution and its smooth and singular components. An appropriate piecewise uniform mesh is constructed, and classical upwind finite difference schemes are used on this mesh to obtain the discrete system of equations. Parameter-uniform error bounds for the numerical approximations are established. Numerical results are provided to illustrate the convergence of the numerical approximations.

• Journal of the Korean Mathematical Society
• /
• v.51 no.4
• /
• pp.679-702
• /
• 2014

Investigation of the numerical solution of singularly perturbed turning point problems dates back to late 1970s. However, due to the presence of layers, not many high order schemes could be developed to solve such problems. On the other hand, one could think of applying the convergence acceleration technique to improve the performance of existing numerical methods. However, that itself posed some challenges. To this end, we design and analyze a novel fitted operator finite difference method (FOFDM) to solve this type of problems. Then we develop a fitted mesh finite difference method (FMFDM). Our detailed convergence analysis shows that this FMFDM is robust with respect to the singular perturbation parameter. Then we investigate the effect of Richardson extrapolation on both of these methods. We observe that, the accuracy is improved in both cases whereas the rate of convergence depends on the particular scheme being used.

• Journal of applied mathematics & informatics
• /
• v.28 no.5_6
• /
• pp.1035-1054
• /
• 2010

We consider a mixed type singularly perturbed one dimensional elliptic problem with discontinuous source term. The domain under consideration is partitioned into two subdomains. A convection-diffusion and a reaction-diffusion type equations are posed on the first and second subdomains respectively. Two hybrid difference schemes on Shishkin mesh are constructed and we prove that the schemes are almost second order convergence in the maximum norm independent of the diffusion parameter. Error bounds for the numerical solution and its numerical derivative are established. Numerical results are presented which support the theoretical results.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.34.1-34
• /
• 2001

We deal with robust control problem for singularly perturbed linear systems with norm-bounded structured uncertainty under state constraints. We assume that the norm-bounded uncertainty is composed of repeated scalar-block and full-block forms. In the structured uncertainty, repeated scalar block forms account for uncertain physical parameter value and full-block forms may be some unknown nonlinear dynamics. In order deal with uncertainty and state constraints, we use LMI(Linear Matrix Inequality). The original problem is decomposed into two well behaved reduced order problems. Shinc two LMI problems are completely independent, each solution can be computed simultaneously and work in parallel.

• Communications of the Korean Mathematical Society
• /
• v.37 no.3
• /
• pp.939-955
• /
• 2022

In this paper, we study a first-order non-linear singularly perturbed Volterra integro-differential equation (SPVIDE). We discretize the problem by a uniform difference scheme on a Bakhvalov-Shishkin mesh. The scheme is constructed by the method of integral identities with exponential basis functions and integral terms are handled with interpolating quadrature rules with remainder terms. An effective quasi-linearization technique is employed for the algorithm. We establish the error estimates and demonstrate that the scheme on Bakhvalov-Shishkin mesh is O(N^-1) uniformly convergent, where N is the mesh parameter. The numerical results on a couple of examples are also provided to confirm the theoretical analysis.

• Journal of applied mathematics & informatics
• /
• v.40 no.1_2
• /
• pp.25-45
• /
• 2022

In this study, the non-standard finite difference method for the numerical solution of singularly perturbed parabolic reaction-diffusion subject to Robin boundary conditions has presented. To discretize temporal and spatial variables, we use the implicit Euler and non-standard finite difference method on a uniform mesh, respectively. We proved that the proposed scheme shows uniform convergence in time with first-order and in space with second-order irrespective of the perturbation parameter. We compute three numerical examples to confirm the theoretical findings.

• Journal of applied mathematics & informatics
• /
• v.40 no.3_4
• /
• pp.491-505
• /
• 2022

This paper is concerned with singularly perturbed convection-diffusion parabolic partial differential equations which have time-delayed. We used the Crank-Nicolson(CN) scheme to build a fitted operator to solve the problem. The underling method's stability is investigated, and it is found to be unconditionally stable. We have shown graphically the unstableness of CN-scheme without fitting factor. The order of convergence of the present method is shown to be second order both in space and time in relation to the perturbation parameter. The efficiency of the scheme is demonstrated using model examples and the proposed technique is more accurate than the standard CN-method and some methods available in the literature, according to the findings.

• Journal of Advanced Information Technology and Convergence
• /
• v.9 no.1
• /
• pp.127-134
• /
• 2019

In terms of deep Reinforcement Learning (RL), exploration can be worked stochastically in the action of a state space. On the other hands, exploitation can be done the proportion of well generalization behaviors. The balance of exploration and exploitation is extremely important for better results. The randomly selected action with ε-greedy for exploration has been regarded as a de facto method. There is an alternative method to add noise parameters into a neural network for richer exploration. However, it is not easy to predict or detect over-fitting with the stochastically exploration in the perturbed neural network. Moreover, the well-trained agents in RL do not necessarily prevent or detect over-fitting in the neural network. Therefore, we suggest a novel design of a deep RL by the balance of the exploration with drop-out to reduce over-fitting in the perturbed neural networks.