하이브리드형 신경망을 이용한 데이터 마이닝의 최적화 기법

강문식,이상용 공주대학교 과학교육연구소 2002 과학교육연구 Vol.33 No.-

최근 데이터 베이스에서 가치있는 정보를 찾아내기 위한 데이터 마이닝 연구가 활발하게 진행되고 있으며, 그 기법으로는 신경망 모델을 사용하는 방법이 많은 연구성과를 거두고 있다. 현재 사용되고 있는 신경망 모델의 문제점은 감독 학습 방법의 경우 목적 패턴의 오류로 유도될 수 있는 오버피팅 문제가 발생하고, 무감독 학습 방법의 경우는 데이터의 정규화 과정으로 인하여 중요한 정보가 왜곡될 수 있어서 만족할 만한 군집화나 분류 효과를 거두지 못하고 있다. 본 논문에서는 이런 문제점들을 해결하기 위해, 하이브리드형 신경망을 제안한다. 실험결과 본 신경망은 어떠한 입력 패턴에 대해 목적 패턴을 만들 수 없거나, 존재하지 않는 경우에 효과적으로 사용될 수 있음을 알 수 있었다. Recently, people have been studying on data mining to find out more valuable information in data bases. There have many good results of studying on the ways to use Neural Network Model. Now, some problems like overfitting are on the rise. Overfitting means that supervised learning methods lead to the error of target patterns. Besides, unsupervised learning methods cause the problem to distort some important information through the process of regularizing data. So we can't expect the satisfying results of clustering and classification. To solve the problems, the paper suggests a hybrid neural networks. According to experiments, the neural networks can be effectively put into use in cases of not being able to make target patterns with some input patterns or non-existing patterns.

ERP시스템의 성공적인 도입 전략

강문식 啓明大學校 産業經營硏究所 1999 經營經濟 Vol.32 No.2

요 약최근에 ERP 패키지의 도입을 통한 기업의 업무 프로세스를 재설계 함으로써 기업의 경쟁력 을 강화하려는 노력이 한창이다. 그러나, 많은 ERP 패키지 도입 시도에도 불구하고 성공사 례가 많지 않은 것도 사실로 드러나고 있다. 이러한 실패 사례는 ERP 패키지에 대한 올바 른 이해 부족으로 인해서 생기는 경우가 많은 것으로 나타나고 있다. 본 연구에서는 ERP 패키지 구축에 영향을 주는 요인들 즉 구축 방법론, 정보시스템, 조직구조, 업무프로세스 및 갈등집단의 관점에서 실패원인을 분석하고 나아가 이를 종합하여 성공적인 구축전략을 제시 한다. ERP도입의 실패 원인은 한 두 가지의 요인에 국한되는 것이 아니고 여러 가지가 복 합적으로 나타난 결과인 것이다. 본 연구에서 제시하는 방법과 전략으로 각 기업에서 ERP 시스템 도입에 대한 분석을 시도함으로써 ERP 시스템의 성공적인 도입에 대한 확신을 얻을 수 있을 것으로 기대한다.ABSTRACTCurrent business environment has been changed so rapidly and companies need to overcome those changes to survive. Many companies have tired to restructure their business operation by employing techniques such as down sizing, BPR, or restructuring. The information systems are the most important parts of the restructuring efforts. Recently, many companies have employed ERP (Enterprise Resources Planning) to solve the problem. However, it is true that not many companies succeed to employ the ERP systems. The failures are often caused by the lack of understanding ERP systems.This paper analyzes the causes of the failures in terms of implementing methods, information systems, organization structures, business processes, and conflicting groups. With the result of the analysis, this study presents the strategies for successful implementation of ERP systems. The failures of employing ERP systems are not caused by one or two factors. They are the result of combined factors. We expect that the strategies we present in this paper with other analysis can assure the successful implementation of ERP systems.

서중조건하에서의 콘크리트 충전 강관 기둥의 시공을 위한 실험적 연구

이장환,강용학,정근호,김진호,최문식,정상진 대한건축학회 2003 대한건축학회 학술발표대회 논문집 - 계획계/구조계 Vol.23 No.1(구조계)

The basic Physical properties, Slump, Slump flow, Air content, Bleeding, and Settlement of concrete was investigated to test Characteristic of Setting and to evaluate the relation between Model specimen and Heat of hydration for construction Over the High Temperature(CFT). The objective of this study is to take the partial core after the cementation of Model specimen, test the compression intensity and analyze the relation to Test piece.

고강도 콘크리트(80MPa이상)를 이용한 콘크리트충전강관구조 적용을 위한 기초물성 연구

이장환,강용학,정근호,김진호,정상진,최문식 대한건축학회 2003 대한건축학회 학술발표대회 논문집 - 계획계/구조계 Vol.23 No.2

One of the most important reasons that CFT is used in many conditions is by using that we can achieve the effect, which reduce the section of the member. This research purpose to find the most ideal composition, which is achieved by the investigation in the concrete's property of matter like ability of Slump, Slump Flow, Air content, Bleeding, and Settlement when the high strength of concrete which is over 80MPa is used m the CFT column.

STAGE EFFICIENCY ESTIMATION BY MODIFIED MIMT USING NLP

Kang, Mun Sik,Kim, In Won,Park, Sun Won 한국화학공학회 1996 Korean Journal of Chemical Engineering Vol.13 No.2

Unmeasured process variables or parameters caused by cost consideration or technical infeasibility can be mostly estimated using data reconciliation techniques. Since, however, the gross errors possibly present in the process measurements deteriorate the data reconciliation results, the reconciled estimates may be biased solutions that are different from the true values. In this paper, the enhanced data reconciliation and gross error detection method, modified MIMT using NLP, was applied to a flash distillation system. It calculated the reconciled values of the measurements as well as the optimal estimates of stage efficiencies which were not measured. These techniques using NLP showed the robustness when compared to the conventional algorithms using linearization techniques.

학술발표회 초록 : ( 공정시스템 (3) - 공정 모델링 및 모사 ) ; 비선형 기법을 이용한 Data Reconciliation 과 Gross Error Detection

강문식(Mun Sik Kang),이우석(Woo Suk Lee),이준석(Jun Seok Lee),김인원(In Won Kim) 한국화학공학회 1994 추계학술발표회 Vol.- No.-

물 전기분해용 SPE-composites의 제조 및 특성 고찰

강문식(Mun Sik Kang),오정훈(Jeong Hun Oh),심규성(Kyu Sung Sim),한학수(Hak Soo Han),설용건(Young Gun Shul),조영일(Yung Il Joe) 한국수소및신에너지학회 1995 한국수소 및 신에너지학회논문집 Vol.6 No.2

SPE-composites which was prepared with impregnation-reduction method for hydroen production were investigated with morphological and electrochemical analysis. As Pt reagent`s concentration increased, Pt loadings increased. However, reductant`s concentration had a little effect on Pt loadings. By the morphological analysis, it was found that Pt was deposited inside of SPE. Furthermore, with electrochemical analysis, we found that Pt loading, electricity, roughness factor and active surface area were 2.05 mg/cm², 14.20 mC, 21.55, 10.51 cm²/mg respectively at reducing agent concentration 0.05 ?/L. Therefore, we found the deposited Pt inside of SPE would give an effect on electrical characteristics of SPE-composites.

Design of the Pohang Accelerator Laboratory (PAL) X-ray Free Electron Laser (XFEL) Test Machine

Mun Gyung Kim,Jinhyuk Choi,Tae Yeon Lee,Heung Sik Kang,Chang-Mook Yim,In Soo Ko 한국물리학회 2008 THE JOURNAL OF THE KOREAN PHYSICAL SOCIETY Vol.53 No.6

The Pohang Accelerator Laboratory is going to build an X-ray free electron laser (FEL) based on self-amplied spontaneous emission (SASE). This new machine will utilize the existing 2.5-GeV injection linac to the storage ring by upgrading its energy up to 3.7 GeV or more. The target wavelength will be 0.3 - 0.45 nm and its third harmonic 0.1 - 0.15 nm will also be used. The project will proceed in two stages: In the first stage, a visible light SASE machine with an energy of 200 MeV will be constructed and tested for the proof of principle. The full X-ray machine will be constructed in the next stage. This paper describes the updated design status of the first-stage test machine. The Pohang Accelerator Laboratory is going to build an X-ray free electron laser (FEL) based on self-amplied spontaneous emission (SASE). This new machine will utilize the existing 2.5-GeV injection linac to the storage ring by upgrading its energy up to 3.7 GeV or more. The target wavelength will be 0.3 - 0.45 nm and its third harmonic 0.1 - 0.15 nm will also be used. The project will proceed in two stages: In the first stage, a visible light SASE machine with an energy of 200 MeV will be constructed and tested for the proof of principle. The full X-ray machine will be constructed in the next stage. This paper describes the updated design status of the first-stage test machine.

Combination effects of tissue heterogeneity and geometric targeting error in stereotactic body radiotherapy for lung cancer using CyberKnife

Kang, Ki Mun,Jeong, Bae Kwon,Choi, Hoon‐,Sik,Yoo, Seung Hoon,Hwang, Ui‐,Jung,Lim, Young Kyung,Jeong, Hojin unknown 2015 Journal of applied clinical medical physics Vol.16 No.5

<P>We have investigated the combined effect of tissue heterogeneity and its variation associated with geometric error in stereotactic body radiotherapy (SBRT) for lung cancer. The treatment plans for eight lung cancer patients were calculated using effective path length (EPL) correction and Monte Carlo (MC) algorithms, with both having the same beam configuration for each patient. These two kinds of plans for individual patients were then subsequently recalculated with adding systematic and random geometric errors. In the ordinary treatment plans calculated with no geometric offset, the EPL calculations, compared with the MC calculations, largely overestimated the doses to PTV by ∼21%, whereas the overestimation were markedly lower in GTV by ∼12% due to relatively higher density of GTV than of PTV. When recalculating the plans for individual patients with assigning the systematic and random geometric errors, no significant changes in the relative dose distribution, except for overall shift, were observed in the EPL calculations, whereas largely altered in the MC calculations with a consistent increase in dose to GTV. Considering the better accuracy of MC than EPL algorithms, the present results demonstrated the strong coupling of tissue heterogeneity and geometric error, thereby emphasizing the essential need for simultaneous correction for tissue heterogeneity and geometric targeting error in SBRT of lung cancer.</P><P>PACS numbers: 87.55.D, 87.55.kh, 87.53.Ly, 87.55.‐x</P>

가공송전 전선 자산데이터의 정제 자동화 알고리즘 개발 연구

Mun, Sung-Duk,Kim, Tae-Joon,Kim, Kang-Sik,Hwang, Jae-Sang 한국전력공사 2021 KEPCO Journal on electric power and energy Vol.7 No.1

As the big data analysis technologies has been developed worldwide, the importance of asset management for electric power facilities based data analysis is increasing. It is essential to secure quality of data that will determine the performance of the RISK evaluation algorithm for asset management. To improve reliability of asset management, asset data must be preprocessed. In particular, the process of cleaning dirty data is required, and it is also urgent to develop an algorithm to reduce time and improve accuracy for data treatment. In this paper, the result of the development of an automatic cleaning algorithm specialized in overhead transmission asset data is presented. A data cleaning algorithm was developed to enable data clean by analyzing quality and overall pattern of raw data.