• 전자공학회논문지C
• /
• 제35C권3호
• /
• pp.79-89
• /
• 1998

The Error Back-Propagation(EBP) algorithm is widely applied to train a multi-layer perceptron, which is a neural network model frequently used to solve complex problems such as pattern recognition, adaptive control, and global optimization. However, the EBP is basically a gradient descent method, which may get stuck in a local minimum, leading to failure in finding the globally optimal solution. Moreover, a multi-layer perceptron suffers from locking a systematic determination of the network structure appropriate for a given problem. It is usually the case to determine the number of hidden nodes by trial and error. In this paper, we propose a new algorithm to efficiently train a multi-layer perceptron. OUr algorithm uses stochastic perturbation in the weight space to effectively escape from local minima in multi-layer perceptron learning. Stochastic perturbation probabilistically re-initializes weights associated with hidden nodes to escape a local minimum if the probabilistically re-initializes weights associated with hidden nodes to escape a local minimum if the EGP learning gets stuck to it. Addition of new hidden nodes also can be viewed asa special case of stochastic perturbation. Using stochastic perturbation we can solve the local minima problem and the network structure design in a unified way. The results of our experiments with several benchmark test problems including theparity problem, the two-spirals problem, andthe credit-screening data show that our algorithm is very efficient.

• 대한건축학회논문집:계획계
• /
• 제35권4호
• /
• pp.127-134
• /
• 2019

Recently, the city development policies are being switched to the policies focusing on recession worldwide. Even though South Korea is also facing the phenomena that small and medium-sized cities are downsized because of a decrease in population and an aging society problem, there is lack of solutions and even they are still planning city development policies based on prospect of high growth. This study is featuring 4 viewpoints depending on shrinking population trend in 7 small and medium sized local cities, which are chosen as downsized cities. It is dealing with population peak, current population, 2030 population, and 2040 population viewpoints. And this study uses downsizing simulation that suggests optimum area that fits population and then shows visual check of the status of the small and medium sized local cities. The area of the small and medium sized local cities are expanded more than needs when they are on population peak point. However, they need only 20% to 25% of current area in 2030 and 2040. The most important thing is realizing the seriousness and facing up to the 'shrinking' phenomena that each small and medium-sized cities suffer. Therefore, this study has significance for presenting the status of small and medium-sized cities' problems in South Korea.

• 한국지능시스템학회논문지
• /
• 제12권4호
• /
• pp.317-322
• /
• 2002

The major operations of Evolutionary Computation include crossover, mutation, competition and selection. Although selection does not create new individuals like crossover or mutation, a poor selection mechanism may lead to problems such as taking a long time to reach an optimal solution or even not finding it at all. In view of this, this paper proposes a hybrid Evolutionary Programming (EP) algorithm that exhibits a strong capability to move toward the global optimum even when stuck at a local minimum using a synergistic combination of the following three basic ideas. First, a "local selection" technique is used in conjunction with the normal tournament selection to help escape from a local minimum. Second, the mutation step has been improved with respect to the Fast Evolutionary Programming technique previously developed in our research group. Finally, the crossover and mutation operations of the Genetic Algorithm have been added as a parallel independent branch of the search operation of an EP to enhance search diversity.

• Journal of Welding and Joining
• /
• 제21권7호
• /
• pp.42-48
• /
• 2003

Welding distortion is a current issue in many industrial parts, especially for heavy industry such as shipbuilding, plant industry. The welding process has many processing parameters influencing welding angular distortion such as heat input power, welding speed, gas flow rate, plate thickness and the welded material properties, etc. In this work, the conventional local minimization concept was applied to find a set of optimum welding process parameters, consisted of welding speed, plate thickness and heat input, for a minimum angular distortion. An analytic solution for welding angular distortion, which is based on laminated plate theory, was also applied to investigate and optimize the welding process parameters. The optimized process parameters and the angular distortion for various parametric conditions could be easily found by using the local minimum concept.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제5권1호
• /
• pp.21-28
• /
• 2005

This paper proposes a new method for accelerating the search speed of genetic algorithms by taking derivative evaluation and conditional random selection into account in their evolution process. Derivative evaluation makes genetic algorithms focus on the individuals whose fitness is rapidly increased. This accelerates the search speed of genetic algorithms by enhancing exploitation like steepest descent methods but also increases the possibility of a premature convergence that means most individuals after a few generations approach to local optima. On the other hand, derivative evaluation under a premature convergence helps genetic algorithms escape the local optima by enhancing exploration. If GAs fall into a premature convergence, random selection is used in order to help escaping local optimum, but its effects are not large. We experimented our method with one combinatorial problem and five complex function optimization problems. Experimental results showed that our method was superior to the simple genetic algorithm especially when the search space is large.

• 한국공작기계학회:학술대회논문집
• /
• 한국공작기계학회 1999년도 추계학술대회 논문집 - 한국공작기계학회
• /
• pp.105-110
• /
• 1999

This paper presents a local obstacle avoidance method for indoor mobile robots using Lane method and velocity Space Command approach. The method locates local obstacles using the information form multi-sensors, such that ultrasonic sensor array and laser scanning sensor. The method uses lane method to determine optimum collision-free heading direction of a robot. Also, it deals with the robot motion dynamics problem to reduce some vibration and guarantee fast movement as well. It yields translational and rotational velocities required to avoid the detected obstacles and to keep the robot heading direction toward goal location as close as possible. For experimental verification of the method, a mobile robot driven by two AC servo motors, equipped with 24 ultrasonic sensor array and laser scanning sensor navigates using the method through a corridor cluttered with obstacle.

• 대한설비공학회:학술대회논문집
• /
• 대한설비공학회 2007년도 동계학술발표대회 논문집
• /
• pp.594-599
• /
• 2007

This study represents numerical analysis on the fluid flow and concentration characteristics by scattering fume at push-pull ventilation system. And the principal point is making optimum on improve an existing ventilation system. This phenomenon simulated about local ventilation system by using commercial CFD tool and base on these fact has find improvements. Advanced model shows most low different pressure and velocity which are suction capability at inlet surface of pull hood has most uniformity. When compared with existing model, Improved model has more good ventilation performance.

• 한국통신학회논문지
• /
• 제32권5B호
• /
• pp.295-303
• /
• 2007

An algorithm is proposed to seek a local optimal solution of the network utility maximization problem in a wireless mesh network, where the architecture being considered is an infrastructure/backbone wireless mesh network. The objective is to achieve proportional fairness amongst the end-to-end flows in wireless mesh networks. In order to establish the communication constraints of the flow rates in the network utility maximization problem, we have presented necessary and sufficient conditions for the achievability of the flow rates. Since wireless mesh networks are generally considered as a type of ad hoc networks, similarly as in wireless multi-hop network, the network utility maximization problem in wireless mesh network is a nonlinear nonconvex programming problem. Besides, the gateway/bridge functionalities in mesh routers enable the integration of wireless mesh networks with various existing wireless networks. Thus, the rate optimization problem in wireless mesh networks is more complex than in wireless multi-hop networks.

• 소음진동
• /
• 제10권5호
• /
• pp.783-791
• /
• 2000

This describes a study on the support location optimizations of the beams using the genetic algorithm and the sensitivity analysis. The genetic algorithm is a probabilistic method searching the optimum at several points simultaneously and requiring only the values of the object and constraint functions. It has therefore more chances to find the global solution and can be applied to the various problems. Nevertheless, it has such a shortcoming that it takes too many calculations, because it is ineffective in local search. While the traditional method using sensitivity analysis is of great advantage in searching the near optimum. thus the combination of the two techniques will make use of the individual advantages, that is, the superiority in global searching form the genetic algorithm and that in local searching form the sensitivity analysis. In this thesis, for the practical applications, the analysis is conducted by FEB ; and as the shapes of structures are taken as the design variation, it requires re-meshing for every analysis. So if it is not properly controlled, the result of the analysis is affected and the optimized solution amy not be the real one. the method is efficiently applied to the problems which the traditional methods are not working properly.

• International Journal of CAD/CAM
• /
• 제7권1호
• /
• pp.51-60
• /
• 2007

This paper presents an algorithm of optimal cutting tool selection for machining of the point-based surface that is defined by a set of surface points rather than parametric polynomial surface equations. As the ball-end and fillet-end mills are generally used for finish machining in a 3-axis computer numerical control machine, the algorithm is applicable for both cutters. The optimum tool would be as large as possible in terms of the cutter radius and/or corner radius which maximise (s) the material removal rate (i.e., minimise (s) the machining time), while still being able to machine the entire point-based surface without gouging any surface point. The gouging are two types: local and global. In this paper, the distance between the cutter bottom and surface points is used to check the local gouging whereas the shortest distance between the surface points and cutter axis is effectively used to check the global gouging. The selection procedure begins with a cutter from the tool library, which has the largest cutter radius and/or corner radius, and then adequacy of the point-density is checked to limit the accuracy of the cutter selection for the point-based surface within tolerance prior to the gouge checking. When the entire surface is gouge-free with a chosen cutting tool then the tool becomes the optimum cutting tool for a list of cutters available in the tool library. The effectiveness of the algorithm is demonstrated considering two examples.