• Journal of the Korean Institute of Intelligent Systems
• /
• 제13권6호
• /
• pp.635-642
• /
• 2003

Reinforcement learning got successful results in a lot of applications such as control and scheduling. Various function approximation methods have been studied in order to improve the learning speed and to solve the shortage of storage in the standard reinforcement learning algorithm of Q-Learning. Most function approximation methods remove some special quality of reinforcement learning and need prior knowledge and preprocessing. Fuzzy Q-Learning needs preprocessing to define fuzzy variables and Local Weighted Regression uses training examples. In this paper, we propose a function approximation method, Fuzzy Q-Map that is based on on-line fuzzy clustering. Fuzzy Q-Map classifies a query state and predicts a suitable action according to the membership degree. We applied the Fuzzy Q-Map, CMAC and LWR to the mountain car problem. Fuzzy Q-Map reached the optimal prediction rate faster than CMAC and the lower prediction rate was seen than LWR that uses training example.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• 제38권3호
• /
• pp.8-13
• /
• 2015

One of the most usual indicators to measure the performance of any inventory policy is the mean physical stock. In general, when estimating the mean physical stock in periodic review inventory systems, approximate approaches are often utilized by practitioners and researchers. The mean physical stock is generally calculated by a simple approximation. Still these simple methods are frequently used to analyze various single stockpoint and multi-echelon inventory systems. However, such a simple approximation can be very inaccurate. This is particularly true for low service levels. Even though exact methods to calculate the mean physical stock have been derived, they are available for specific cases only and computationally not very efficient, and therefore less useful in practice. In literature, approximate approaches, such as the simple, the linear, and Simpson approximations, were derived for the periodic review inventory systems that allow backorders. This paper modifies the approximate approaches for the lost sales case and evaluates the modified approximate approaches. Through computational experiments, average (and maximum) percentage deviations of mean physical stock between the exact method and the modified approximations are compared in the periodic review inventory system with lost sales. The same comparison between the modified and the original approximations are also conducted, in order to examine the performance of modified approximations. The results show that all modified approximations perform well for high service levels, but also that the performance may deteriorate fast with decreasing service level. The modified Simpson approximation is clearly better. In addition, the comparison between the modified and the original approximations in the periodic review inventory system with lost sales shows that the modified approximation outperforms the original approximation.

• Journal of the Korean Operations Research and Management Science Society
• /
• 제27권4호
• /
• pp.41-65
• /
• 2002

We study a multi-component production/inventory system in which individual components are made to meet various demand types. We assume that the demands arrive according to a Poisson process, but there is a fixed probability that a demand requests a particular kit of different components. Each component is produced by a flow line with several stations in which the processing times of each station follow a two-stage Coxian distribution. The production of each component is operated by an independent base-stock policy with blocking. We assume that the time needed to assemble final products follows a general distribution and the capacity of an assembling facility is sufficiently large. The objective of this study is to obtain key performance measures such as the distribution of the number of each orders for each final product and the mean time of fulfilling a customer order. The basic principle of the proposed approximation method is to decompose the original system into a set of subsystems, each subsystem being associated with a flow line. Each subsystem is analyzed in isolation using a Marie's method. An iterative procedure is then used to determine the unknown parameters of each subsystem. Numerical results show that the accuracy of the approximation method is acceptable.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• 제31권4호
• /
• pp.134-139
• /
• 2008

Most approaches for continuous review inventory problem need tables for loss function and cumulative standard normal distribution. Furthermore, it is time-consuming to calculate order quantity (Q) and reorder point (r) iteratively until required values are converged. The purpose of this paper is to develop a direct method to get the solution without any tables. We used approximation approaches for loss function and cumulative standard normal distribution. The proposed method can get the solution directly without any iterative procedure for Q, r and without any tables. The performance of the proposed approach is tested by using numerical examples. The budget constraint of this paper assumes that purchasing costs are paid at the time an order is arrived. This constraint can be easily replaced by capacity constraint or budget constraint in which' purchasing costs are paid at the time an order is placed.

• Communications for Statistical Applications and Methods
• /
• 제3권3호
• /
• pp.93-101
• /
• 1996

The unknown response function is usually approximated by a low order polynomial model. Such an approximation always accompanies bias due to model departure. The minimax Average MSE (AMSE) designs are suggested for estimating mean responses. A class of first order minimax AMSE designs is derived and a specific first order minimax AMSE design is selected from the class by optimizing the secondary criterion related to the power of the lack of fit test.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• 제4권1호
• /
• pp.67-77
• /
• 2000

In this paper we consider finite element mathods for nonlinear parabolic problems defined in ${\Omega}{\subset}R^d$ ($d{\leq}4$). A new initial approximation is taken. Optimal order error estimates in $L_p$ for $2{\leq}p{\leq}{\infty}$ are established for arbitrary order finite element. One order superconvergence in $W^{1,p}$ for $2{\leq}q{\leq}{\infty}$ are demonstrated as well.

• Communications for Statistical Applications and Methods
• /
• 제6권1호
• /
• pp.305-318
• /
• 1999

In this paper we derive the higher order expansions of the first four cumulants of multi-dimensional Maximum Likelihood Estimator (MLE) under the general parametric model up to and including terms of order O({{{{ {n }^{-1 } }}}}) Also we obtain the explicit form of the expansion of the normalizing trans formation of multi-dimensional MLE and show that the suggested normalizing process is much better than the normal approximation based on central limit theorem through example.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2000년도 제15차 학술회의논문집
• /
• pp.12-12
• /
• 2000

This paper consider a new weighted model reduction using block diagonal solutions of Lyapunov inequalities. With the input and/or output weighting function, the stability of reduced order system is quaranteed and a priori error bound is proposed. to achieve this, after finding the solutions of two Lyapunov inequalities and balancing the full order system, we find the reduced order systems using the direct truncation and the singular perturbation approximation. The proposed method is compared with other existing methods using numerical example.

• Communications for Statistical Applications and Methods
• /
• 제26권3호
• /
• pp.305-313
• /
• 2019

Generalized canonical correlation analysis (GCCA) extends the canonical correlation analysis (CCA) to the case of more than two sets of variables and there have been many studies on how two-set canonical solutions can be generalized. In this paper, we derive certain stationary equations which can lead the higher-order solutions of several GCCA methods and suggest a type of iterative procedure to obtain the canonical coefficients. In addition, with some numerical examples we present the methods for graphical display, which are useful to interpret the GCCA results obtained.

• Journal of the Earthquake Engineering Society of Korea
• /
• 제10권2호
• /
• pp.21-29
• /
• 2006

It is impractical to install sensors on every floor of a tall building to measure the full state vector because of the large number of degrees of freedom. This makes it necessary to introduce reduced order control. A kind of system reduction scheme (dynamic condensation method) is proposed in this paper. This method is iterative and Guyan condensation is looked upon as an initial approximation of the iteration. Since the reduced order system is updated repeatedly until a desired one is obtained, the accuracy of the reduced order system resulting from the proposed method is much higher than that obtained from the Guyan condensation method. An eigenvalue shilling technique is applied to accelerate the convergence of Iteration. Two schemes to establish the reduced order system by using the proposed method are also presented and discussed in this paper. The results for a tail building with active tuned mass damper show that the proposed method is efficient for the reduced order modelling and the accuracy is very close to exact only after two iterations.