Simulation Optimization with Statistical Selection Method

김주미 한국경영과학회 2007 Management Science and Financial Engineering Vol.13 No.1

I propose new combined randomized methods for global optimization problems. These methods are based on the Nested Partitions (NP) method, a useful method for simulation optimization which guarantees global optimal solution but has several shortcomings. To overcome these shortcomings I hired various statistical selection methods and combined with NP method. I first explain the NP method and statistical selection method. And after that I present a detail description of proposed new combined methods and show the results of an application. As well as, I show how these com-bined methods can be considered in case of computing budget limit problem.

Feature Selection for Steel Defects Classification

Daun Jeong,Dongyeop Kang,Sangchul Won 제어로봇시스템학회 2010 제어로봇시스템학회 국제학술대회 논문집 Vol.2010 No.10

In this paper, features of steel defects data are selected using a wrapper algorithm to increase classification performance. The data are constructed using images of steel defects which are classified two classes as defects and pseudo defects. The suggested algorithm selects features which are relevant to class using the kappa statistic. This measure is suggested to improve accuracy of minor class because steel defects data are highly imbalanced. The several algorithms were compared with the algorithm to show performances.

Characteristics of a plasma information variable in phenomenology-based, statistically-tuned virtual metrology to predict silicon dioxide etching depth

장윤창,노현준,박설혜,정상민,Sanywon Ryu,권지원,김남균,김곤호 한국물리학회 2019 Current Applied Physics Vol.19 No.10

A phenomenology-based virtual metrology (VM) for monitoring SiO2 etching depth was proposed by Park (2015). It achieved high prediction accuracy by introducing newly developed plasma information (PI) variables as designated inputs, called PI-VM. The PI variables represent the state of the plasma, the sheath, and the target during the process. We investigate how a PI variable can help to improve prediction accuracy of VM and how it plays a special role in the statistical selection. We choose only PIEEDF among the three PI variables to focus on the investigation. The PIEEDF is determined from the ratio of line-intensities of optical emission spectroscopy. We apply Pearson's correlation filter (PCF), principal component analysis (PCA), and stepwise variable selection (SVS) as statistical selection methods on the variables set including PIEEDF or not. Multilinear regression is used to model the VM. This study reveals that PIEEDF variable is a good variable in terms of independence from other input variables and explanatory power for an output variable. Especially, VM using SVS method applied to variable sets including PIEEDF achieves the highest accuracy, comparable to Park's PI-VM. This study shows that PIEEDF variable is particularly useful for monitoring of the fine variations in semiconductor manufacturing process and it also extends the utilization of OES sensor data.

Characteristics of a plasma information variable in phenomenology-based, statistically-tuned virtual metrology to predict silicon dioxide etching depth

Jang, Yunchang,Roh, Hyun-Joon,Park, Seolhye,Jeong, Sangmin,Ryu, Sanywon,Kwon, Ji-Won,Kim, Nam-Kyun,Kim, Gon-Ho Elsevier 2019 CURRENT APPLIED PHYSICS Vol.19 No.10

<P><B>Abstract</B></P> <P>A phenomenology-based virtual metrology (VM) for monitoring SiO<SUB>2</SUB> etching depth was proposed by Park (2015). It achieved high prediction accuracy by introducing newly developed plasma information (PI) variables as designated inputs, called PI-VM. The PI variables represent the state of the plasma, the sheath, and the target during the process. We investigate how a PI variable can help to improve prediction accuracy of VM and how it plays a special role in the statistical selection. We choose only PI<SUB>EEDF</SUB> among the three PI variables to focus on the investigation. The PI<SUB>EEDF</SUB> is determined from the ratio of line-intensities of optical emission spectroscopy. We apply Pearson's correlation filter (PCF), principal component analysis (PCA), and stepwise variable selection (SVS) as statistical selection methods on the variables set including PI<SUB>EEDF</SUB> or not. Multilinear regression is used to model the VM. This study reveals that PI<SUB>EEDF</SUB> variable is a good variable in terms of independence from other input variables and explanatory power for an output variable. Especially, VM using SVS method applied to variable sets including PI<SUB>EEDF</SUB> achieves the highest accuracy, comparable to Park's PI-VM. This study shows that PI<SUB>EEDF</SUB> variable is particularly useful for monitoring of the fine variations in semiconductor manufacturing process and it also extends the utilization of OES sensor data.</P>

통계적 여과 기법 기반 무선 센서 네트워크의 영역별 경로 설정 방법

박혁,문수영,조대호 한국정보통신설비학회 2010 한국정보통신설비학회 학술대회 Vol.2010 No.1

통계적 방법론을 이용한 호우피해예측함수 개발

최창현(Choi, Changhyun),김종성(Kim, Jongsung),김정환(Kim, Jeonghwan),김한용(Kim, Hanyong),이우주(Lee, Woojoo),김형수(Kim, Hung Soo) 한국방재학회 2017 한국방재학회논문집 Vol.17 No.3

본 연구에서는 한강권역을 대상으로 선형회귀모형, 일반화선형모형, 주성분 회귀모형, 인공신경망 모형과 같은 통계적 모형을 적용하여 호우피해예측함수를 개발하였다. 학습용 데이터(1994∼2011년)로부터 개발된 함수를 평가용 데이터(2012∼2015년)에 적용하고, 실제 호우피해액과 예측 호우 피해액을 비교하여 예측력을 평가하였다. 평가결과 NRMSE는 10.61∼13.89%로 나타났으며, 일반화선형모형에 벌점화를 통한 축소추정법을 적용한 함수에서 가장 좋은 예측력을 나타냈다. 본 연구에서 개발된호우피해예측함수를 활용하여 재난 피해 발생 전 피해규모와 영향을 신속하게 추정한다면, 예방 및 대비 차원의 재난관리에 유용하게 활용될 수 있을 것이다. In this study, we develop heavy rain damage prediction functions for Han river basin by using statistical models such as linear regression model, generalized linear model, principal component regression model, artificial neural network model. The prediction functions were estimated from the training data (1994 to 2011) and evaluated by the test data (2012 to 2015). Their performances were assessed by comparing observed heavy rain damages and predicted damages. Specifically, the NRMSE was 10.61~13.89%. A generalized linear model based on penalized likelihood method showed the best prediction performance. This heavy rain damage prediction function developed in this study can be used not only for estimati

Introduction to the production procedure of representative annual maximum precipitation scenario for different durations based on climate change with statistical downscaling approaches

Lee Taesam 한국수자원학회 2018 한국수자원학회논문집 Vol.51 No.11

기후변화는 홍수의 가장 큰 원인이 되는 극치강우의 빈도와 크기에 매우 큰 영향을 미치고 있다. 특히, 우리나라에서 발생하는 대규모 재해는 강우에 의한 홍수피해가 대부분을 차지하고 있다. 이러한 홍수피해는 기후변화에 의한 극한강우의 발생 빈도가 높아짐에 따라 새로운 재해양상으로 전개되고 있다. 하지만, 미래 기후변화 시나리오 자료는 해상도의 한계로 인하여 중소규모 하천 및 도시유역에 요구되는 수준의 자료 수집이 불가능한 상태이다. 이러한 문제점을 개선하기 위하여 본 연구에서는 전지구모형에서 생산된 기후변화 시나리오에 대해서 여러 단계의 통계적 상세화 기법을 통하여 우리나라 전역에 대하여 미래 시나리오에 대한 빈도해석이 가능하도록 각 지점의 특성에 따라 시간적으로 상세화하기 위해 개발된 방 법 및 과정을 소개하였다. 이를 통해, 시간상세화 자료를 토대로 미래 강우에 대한 빈도해석과 기후변화에 따른 방재성능 목표강우량을 산정하는데 활용할 수 있도록 하였다. Climate change has been influenced on extreme precipitation events, which are major driving causes of flooding. Especially, most of extreme water-related disasters in Korea occur from floods induced by extreme precipitation events. However, future climate change scenarios simulated with Global Circulation Models (GCMs) or Reigonal Climate Models (RCMs) are limited to the application on medium and small size rivers and urban watersheds due to coarse spatial and temporal resolutions. Therefore, the current study introduces the state-of-the-art approaches and procedures of statistical downscaling techniques to resolve this limitation It is expected that the temporally downscaled data allows frequency analysis for the future precipitation and estimating the design precipitation for disaster prevention.

통계적 상세화 기법을 통한 기후변화기반 지속시간별 연최대 대표 강우시나리오 생산기법 소개

이태삼 한국수자원학회 2018 한국수자원학회논문집 Vol.51 No.S-1

기후변화는 홍수의 가장 큰 원인이 되는 극치강우의 빈도와 크기에 매우 큰 영향을 미치고 있다. 특히, 우리나라에서 발생하는 대규모 재해는 강우에 의한 홍수피해가 대부분을 차지하고 있다. 이러한 홍수피해는 기후변화에 의한 극한강우의 발생 빈도가 높아짐에 따라 새로운 재해양상으로 전개되고 있다. 하지만, 미래 기후변화 시나리오 자료는 해상도의 한계로 인하여 중소규모 하천 및 도시유역에 요구되는 수준의 자료 수집이 불가능한 상태이다. 이러한 문제점을 개선하기 위하여 본 연구에서는 전지구모형에서 생산된 기후변화 시나리오에 대해서 여러 단계의 통계적 상세화 기법을 통하여 우리나라 전역에 대하여 미래 시나리오에 대한 빈도해석이 가능하도록 각 지점의 특성에 따라 시간적으로 상세화하기 위해 개발된 방법 및 과정을 소개하였다. 이를 통해, 시간상세화 자료를 토대로 미래 강우에 대한 빈도해석과 기후변화에 따른 방재성능 목표강우량을 산정하는데 활용할 수 있도록 하였다. Climate change has been influenced on extreme precipitation events, which are major driving causes of flooding. Especially, most of extreme water-related disasters in Korea occur from floods induced by extreme precipitation events. However, future climate change scenarios simulated with Global Circulation Models (GCMs) or Reigonal Climate Models (RCMs) are limited to the application on medium and small size rivers and urban watersheds due to coarse spatial and temporal resolutions. Therefore, the current study introduces the state-of-the-art approaches and procedures of statistical downscaling techniques to resolve this limitation It is expected that the temporally downscaled data allows frequency analysis for the future precipitation and estimating the design precipitation for disaster prevention.

영어 동사의 의미적 유사도와 논항 선택 사이의 연관성: ICE-GB와 WordNet을 이용한 통계적 검증

송상헌 ( Sang Houn Song ),최재웅 ( Jae Woong Choe ) 한국언어정보학회 2010 언어와 정보 Vol.14 No.1

플라즈마 정보인자를 활용한 SiO₂ 식각 깊이 가상 계측 모델의 특성 인자 역할 분석

장윤창,박설혜,정상민,유상원,김곤호,Jang, Yun Chang,Park, Seol Hye,Jeong, Sang Min,Ryu, Sang Won,Kim, Gon Ho 한국반도체디스플레이기술학회 2019 반도체디스플레이기술학회지 Vol.18 No.4

We analyzed how the features in plasma information based virtual metrology (PI-VM) for SiO2 etching depth with variation of 5% contribute to the prediction accuracy, which is previously developed by Jang. As a single feature, the explanatory power to the process results is in the order of plasma information about electron energy distribution function (PIEEDF), equipment, and optical emission spectroscopy (OES) features. In the procedure of stepwise variable selection (SVS), OES features are selected after PIEEDF. Informative vector for developed PI-VM also shows relatively high correlation between OES features and etching depth. This is because the reaction rate of each chemical species that governs the etching depth can be sensitively monitored when OES features are used with PIEEDF. Securing PIEEDF is important for the development of virtual metrology (VM) for prediction of process results. The role of PIEEDF as an independent feature and the ability to monitor variation of plasma thermal state can make other features in the procedure of SVS more sensitive to the process results. It is expected that fault detection and classification (FDC) can be effectively developed by using the PI-VM.