• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• v.9 no.1
• /
• pp.479-484
• /
• 2005

본 논문에서는 오류 역전파 알고리즘에서 은닉층의 노드 수를 설정하는 문제와 ART1의 경계 변수의 설정에 따른 인식률이 저하되는 문제점을 개선하기 위해 ART1 알고리즘과 퍼지 단층 지도 학습 알고리즘을 결합한 ART1 기반 퍼지 지도 학습 알고리즘을 제안한다. 제안된 알고리즘은 가중치 조정에 승자 뉴런 방식을 도입하여 은닉층에 해당하는 클래스에 영향을 끼친 패턴들의 정보만 저장하게 하여 은닉층 노드로의 책임 분담에 의한 정체 현상이 일어날 가능성을 줄인다. 그리고 학습시간과 학습의 수렴성도 개선한다. 제안된 알고리즘의 학습 성능을 분석하기 위하여 주민등록번호 분류를 대상으로 실험한 결과, 제안된 방법이 기존의 신경망보다 경계 변수나 모멘트에 민감하지 않으며 학습 시간도 적게 소요되고 수렴성도 우수한 성능이 있음을 확인하였다.

• Journal of the Korean Society for Precision Engineering
• /
• v.27 no.6
• /
• pp.64-74
• /
• 2010

This paper proposes a new TR-based baseline-free SHM technique in which the time-reversal (TR) property of the guided Lamb waves is utilized. The new TR-based SHM technique has two distinct features when compared with the other TR-based SHM techniques: (1) The backward TR process commonly conducted by the measurement is replaced by the computation-based process; (2) In place of the comparison method, the TOF information of the damage signal extracted from the reconstructed signal is used for the damage diagnosis in conjunction with the imaging method which enables us to represent the damage as an image. The proposed TR-based SHM technique is then validated through the damage diagnosis experiment for an aluminum plate with a damage at different locations.

• Proceedings of the Korean Magnestics Society Conference
• /
• 2002.04a
• /
• pp.74-75
• /
• 2002

• Journal of the Korean Magnetics Society
• /
• v.11 no.6
• /
• pp.262-266
• /
• 2001

The magnetic switching volume is known as an important parameter to understand the magnetization reversal process, thermal stability of the written information and media noise. This parameter is influenced significantly by the microstructure of the magnetic layer as well as underlayer. Therefore, we fabricated CoSm/Cr thin films with varying magnetic layer thickness under constant sputtering by using a dc magnetic sputtering machine. The magnetic layer thickness effect on the magnetic switching volume have been studied by the means of magnetic viscosity and dc demagnetization remanence curve mesurements. From these measurements, we found that the switching volumes increased with increasing the magnetic layer thickness, whereas the coercivity showed different behavior. These may be a result of the increased intergranular coupling and the larger volume fraction of the magnetic layer.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.7
• /
• pp.1759-1769
• /
• 1997

The Error BackPropagation (EBP) algorithm for multi-layered neural networks is widely used in various areas such as associative memory, speech recognition, pattern recognition and robotics, etc. Nevertheless, many researchers have continuously published papers about improvements over the original EBP algorithm. The main reason for this research activity is that EBP is exceeding slow when the number of neurons and the size of training set is large. In this study, we developed new learning speed acceleration methods using variable learning rate, variable momentum rate and variable slope for the sigmoid function. During the learning process, these parameters should be adjusted continuously according to the total error of network, and it has been shown that these methods significantly reduced learning time over the original EBP. In order to show the efficiency of the proposed methods, first we have used binary data which are made by random number generator and showed the vast improvements in terms of epoch. Also, we have applied our methods to the binary-valued Monk's data, 4, 5, 6, 7-bit parity checker and real-valued Iris data which are famous benchmark training sets for machine learning.

• Korean Journal of Agricultural and Forest Meteorology
• /
• v.19 no.3
• /
• pp.75-85
• /
• 2017

A method to estimate hourly temperature profiles on calm and clear nights was developed based on temporal changes of inversion height and strength. A meteorological temperature profiler (Model MTP5H, Kipp and Zonen) was installed on the rooftop of the Highland Agriculture Research Institute, located in Daegwallyeong-myeon, Pyeongchang-gun, Gangwon-do. The hourly vertical distribution of air temperature was measured up to 600 m at intervals of 50 m from May 2007 to March 2008. Temperature and relative humidity data loggers (HOBO U23 Pro v2, Onset Computer Corporation, USA) were installed in the Jungdae-ri Valley, located between Gurye-gun, Jeollanam-do and Gwangyang-si, Jeollanam-do. These loggers were used to archive measurements of weather data 1.5 m above the surface from October 3, 2014, to November 23, 2015. The inversion strength was determined using the difference between the temperature at the inversion height, which is the highest temperature in the profile, and the temperature at 100 m from the surface. Empirical equations for the changes of inversion height and strength were derived to express the development of temperature inversion on calm and clear nights. To estimate air temperature near the ground on a slope exposed to crops, the equation's parameters were modified using temperature distribution of the mountain slope obtained from the data loggers. Estimated hourly temperatures using the method were compared with observed temperatures at 19 weather sites located within three watersheds in the southern Jiri-mountain in 2015. The mean error (ME) and root mean square error (RMSE) of the hourly temperatures were $-0.69^{\circ}C$ and $1.61^{\circ}C$, respectively. Hourly temperatures were often underestimated from 2000 to 0100 LST the next day. When temperatures were estimated at 0600 LST using the existing model, ME and RMSE were $-0.86^{\circ}C$ and $1.72^{\circ}C$, respectively. The method proposed in this study resulted in a smaller error, e.g., ME of $-0.12^{\circ}C$ and RMSE of $1.34^{\circ}C$. The method could be improved further taking into account various weather conditions, which could reduce the estimation error.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2005.05b
• /
• pp.904-907
• /
• 2005

본 연구에서는 강우량의 실측치인 자동기상관측소(AWS) 자료와 현재의 대기상태인 인공위성(GMS-5호) 자료를 입력자료로 하여 현재부터 3시간 선행시간까지의 면적평균강우량을 예측할 수 있도록 강우예측 신경망 모형을 개발하였으며, 2002년 8월 집중호우시 남강댐 유역에 적용하였다. 신경망 모형의 학습을 위해서 $1998\~2001$년 $6\~9$월과 2002년 6, 7월의 강우사상과 적외선 자료가 사용되었고, 학습이 종료되면 예측기간(2002년 8월 $6\~16$일)동안의 강우예측이 수행되었다. 신경망 모형의 학습단계에서는 자료들간의 비선형 상관관계를 나타내는데 적합한 역전파 알고리즘 학습방법 중 모멘텀법을 사용하였으며, 신경망 모형의 출력값은 현재부터 3시간 후까지의 면적평균강우량을 예측할 수 있도록 구성하였다. 예측된 면적평균강우량은 실제 관측된 강우량의 패턴은 잘 따르고 있었지만 첨두치를 과소평가하는 경향이 나타났다. 본 연구에서 개발된 신경망 모형은 관측된 강우자료의 품질과 패턴이 모형의 정확성에 미치는 영향이 절대적인 기존의 신경망 모형과 차별화하여, 현재의 대기상태를 나타내는 인공위성 자료를 추가함으로써 보다 정확한 강우량 예측이 가능하도록 하였다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2008.05a
• /
• pp.1027-1031
• /
• 2008

본 연구는 한반도 영역을 대상으로 2001년 7, 8월과 2002년 6월로 홍수기를 대상으로 RDAPS 모형, AWS, 상층기상관측(upper-air sounding)의 자료를 이용하였다. 또한 수치예보자료를 범주적 예측확률로 변환하고 인공신경망기법(ANN)을 이용하여 강수발생확률의 예측정확성을 향상시키는데 있다. 신경망의 예측인자로 사용된 대기변수는 500/ 750/ 1000hpa에서의 지위고도, 500-1000hpa에서의 층후(thickness), 500hpa에서의 X와 Y의 바람성분, 750hpa에서의 X와 Y의 바람성분, 표면풍속, 500/ 750hpa/ 표면에서의 온도, 평균해면기압, 3시간 누적 강수, AWS관측소에서 관측된 RDAPS모형 실행전의 6시간과 12시간동안의 누적강수, 가강수량, 상대습도이며, 예측변수로는 강수발생확률로 선택하였다. 강우는 다양한 대기변수들의 비선형 조합으로 발생되기 때문에 예측인자와 예측변수 사이의 복잡한 비선형성을 고려하는데 유용한 인공신경망을 사용하였다. 신경망의 구조는 전방향 다층퍼셉트론으로 구성하였으며 역전파알고리즘을 학습방법으로 사용하였다. 강수예측성과의 질을 평가하기 위해서 $2{\times}2$ 분할표를 이용하여 Hit rate, Threat score, Probability of detection, Kuipers Skill Score를 사용하였으며, 신경망 학습후의 강수발생확률은 학습전의 강수발생확률에 비하여 한반도영역에서 평균적으로 Kuipers Skill Score가 0.2231에서 0.4293로 92.39% 상승하였다.

• Korean Journal of Cognitive Science
• /
• v.4 no.2
• /
• pp.181-200
• /
• 1994

In the learning phase of multilyer feedforword neural network,there are problems such that local minimum,learning praralysis and slow learning speed when backpropagation algorithm used.To overcome these problems, the genetic algorithm has been used as learing method in the multilayer feedforword neural network instead of backpropagation algorithm.However,because the genetic algorith, does not have any mechanism for fine-tuned local search used in backpropagation method,it takes more time that the genetic algorithm converges to a global optimal solution.In this paper,we suggest a new GA-BP method which provides a fine-tunes local search to the genetic algorithm.GA-BP method uses gradient descent method as one of genetic algorithm's operators such as mutation or crossover.To show the effciency of the developed method,we applied it to the 3-parity bit problem with analysis.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.1
• /
• pp.21-31
• /
• 2000

In this paper, an analysis of the stability for a class of discrete-time neural network control systems is presentd. Based on Lyapunov's direct method, a sufficient stability condition for the neural network control systems is systematically derived and the modified back propagation algorithm which reflects the derived stability condition is suggested. The modified BP originates from the derived sufficient condition and guarantees the exponential stability of the resulting trained closed system. Finally, computer simulation is included to show an example where the derived stability condition and the BP modified bythe condition is used to train the control plant.