• Proceedings of the KIPE Conference
• /
• 2012.07a
• /
• pp.540-541
• /
• 2012

본 논문에서는 신재생에너지원인 1.5MW 풍력발전기가 연계된 계통에 2MVA/500kWh BESS(Battery Energy Storage System)를 적용하여 실증시험을 수행한 결과를 기술하였다. 풍력발전기의 출력 전력을 측정하여 제어 알고리즘에 따라 충, 방전 지령값을 계산하는 상위제어기 EMS와 BESS를 연동하여 운전하였다. 이를 통해 BESS를 이용하여 풍력발전기의 출력이 심하게 변동하여도 계통으로 송전되는 전력을 안정적으로 제어할 수 있음을 검증하였다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06c
• /
• pp.317-319
• /
• 2012

본 논문에서는 아웃라이어 처리를 통한 강인한 HDR 영상 복원 방법을 제시한다. 기존의 방법들은 LDR 영상들에서 흔히 발생하는 긴 노출시간으로 인한 블러 현상이나 저노출/과노출로 인한 포화 픽셀(아웃라이어)을 고려하지 않았다. 본 논문이 제시하는 방법은 MAP(Maximum a priori)을 이용하여 블러 및 아웃라이어를 반영하여 HDR 영상 복원 문제를 정확히 모델링하고, 블러 추정 및 EM(Expectation-Maximization) 알고리즘 기반의 아웃라이어 추정을 통해 품질 저하가 없는 선명한 HDR 영상을 복원한다. 실험 결과를 통해 본 논문이 제시하는 방법이 블러 및 아웃라이어를 포함하는 LDR 영상들로부터 우수한 품질의 HDR 영상을 효과적으로 복원할 수 있음을 보이며, 최근에 개발된 방법들과 비교해서도 더 우수한 품질을 갖는 것을 볼 수 있다.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.459-461
• /
• 2020

매일 게시되는 다양한 프로야구 관련 기사에는 경기 결과, 각종 기록, 선수의 부상 등 다양한 정보가 뒤섞여있어, 사용자가 원하는 정보를 찾아내는 과정이 매우 번거롭다. 본 논문에서는 문서 검색과 기계 독해를 이용하여 야구 분야에 대한 Q&A 시스템을 제안한다. 기사를 형태소 분석하고 BM25 알고리즘으로 얻은 문서 가중치로 사용자 질의에 적합한 기사들을 선정하고 KorQuAD 1.0과 직접 구축한 프로야구 질의응답 데이터셋을 이용해 학습시킨 BERT 모델 기반 기계 독해로 답변 추출을 진행한다. 야구 특화 데이터 셋을 추가하여 학습시켰을 때 F1 score, EM 모두 15% 내외의 정확도 향상을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.335-339
• /
• 2021

기계독해(MRC)는 인공지능 알고리즘이 문서를 이해하고 질문에 대한 정답을 찾는 기술이다. MRC는 사전 학습 모델을 사용하여 높은 성능을 내고 있고, 일반 텍스트문서 뿐만 아니라 문서 내의 테이블(표)에서도 정답을 찾고자 하는 연구에 활발히 적용되고 있다. 본 연구에서는 기존의 사전학습 모델을 테이블 데이터에 활용하여 질의응답을 할 수 있는 방법을 제안한다. 더불어 테이블 데이터를 효율적으로 학습하기 위한 데이터 구성 방법을 소개한다. 사전학습 모델은 BERT[1]를 사용하여 테이블 정보를 인코딩하고 Masked Entity Recovery(MER) 방식을 사용한다. 테이블 질의응답 모델 학습을 위해 한국어 위키 문서에서 표와 연관 텍스트를 추출하여 사전학습을 진행하였고, 미세 조정은 샘플링한 테이블에 대한 질문-답변 데이터 약 7만건을 구성하여 진행하였다. 결과로 KorQuAD2.0 데이터셋의 테이블 관련 질문 데이터에서 EM 69.07, F1 78.34로 기존 연구보다 우수한 성능을 보였다.

• Journal of the Korean Data and Information Science Society
• /
• v.26 no.4
• /
• pp.885-894
• /
• 2015

It is necessary to forecast the electricity demand for reliable and effective operation of the power system. In this study, we try to categorize a functional data, the mean curve in accordance with the time of daily power demand pattern. The data were collected between January 1, 2009 and December 31, 2011. And it were converted to time series data consisting of seasonal components and error component through log transformation and removing trend. Functional clustering by Ma et al. (2006) are applied and parameters are estimated using EM algorithm and generalized cross validation. The number of clusters is determined by classifying holidays or weekdays. Monday, weekday (Tuesday to Friday), Saturday, Sunday or holiday and season are described the mean curve of daily power demand pattern.

• The Korean Journal of Applied Statistics
• /
• v.24 no.6
• /
• pp.987-994
• /
• 2011

From the point view of credit evaluation whose population is divided into the default and non-default state, two methods are considered to estimate conditional distribution functions: one is to estimate under the assumption that the data is followed the mixture normal distribution and the other is to use the kernel density estimation. The parameters of normal mixture are estimated using the EM algorithm. For the kernel density estimation, five kinds of well known kernel functions and four kinds of the bandwidths are explored. In addition, the corresponding ROC functions are obtained based on the estimated distribution functions. The goodness-of-fit of the estimated distribution functions are discussed and the performance of the ROC functions are compared. In this work, it is found that the kernel distribution functions shows better fit, and the ROC function obtained under the assumption of normal mixture shows better performance.

• Communications for Statistical Applications and Methods
• /
• v.17 no.5
• /
• pp.689-696
• /
• 2010

Interval-censored data are commonly found in studies of diseases that progress without symptoms, which require clinical evaluation for detection. Several techniques have been suggested with independent assumption. However, the assumption will not be valid if observations come from clusters. Furthermore, when the cluster size relates to response variables, commonly used methods can bring biased results. For example, in a study on lymphatic filariasis, a parasitic disease where worms make several nests in the infected person's lymphatic vessels and reside until adulthood, the response variable of interest is the nest-extinction times. Since the extinction times of nests are checked by repeated ultrasound examinations, exact extinction times are not observed. Instead, data are composed of two examination points: the last examination time with living worms and the first examination time with dead worms. Furthermore, as Williamson et al. (2008) pointed out, larger nests show a tendency for low clearance rates. This association has been denoted as an informative cluster size. To analyze the relationship between the numbers of nests and interval-censored nest-extinction times, this study proposes a joint model for the relationship between cluster size and clustered interval-censored failure data.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.1
• /
• pp.11-18
• /
• 2002

In this paper, a gain-adapted speech recognition method in noise is developed in the time domain. Noise is assumed to be colored. To cope with the notable nonstationary nature of speech signals such as fricative, glides, liquids, and transition region between phones, the nonstationary autoregressive (NAR) hidden Markov model (HMM) is used. The nonstationary AR process is represented by using polynomial functions with a linear combination of M known basis functions. When only noisy signals are available, the estimation problem of noise inevitably arises. By using multiple Kalman filters, the estimation of noise model and gain contour of speech is performed. Noise estimation of the proposed method can eliminate noise from noisy speech to get an enhanced speech signal. Compared to the conventional ARHMM with noise estimation, our proposed NAR-HMM with noise estimation improves the recognition performance about 2-3%.

• Journal of Korea Multimedia Society
• /
• v.15 no.6
• /
• pp.748-760
• /
• 2012

In case brightness distribution is concentrated in a region, it is difficult to classify the image features. To solve this problem, we apply global histogram equalization and local histogram equalization to images. In case of global histogram equalization, it can be too bright or dark because it doesn't consider the density of brightness distribution. Thus, it is difficult to enhance the local contrast in the images. In case of local histogram equalization, it can produce unexpected blocks in the images. In order to enhance the contrast in the images, this paper proposes a local histogram equalization based on the Gaussian Mixture Models(GMMs) in regions of histogram. Mean and variance parameters in each regions is updated EM-algorithm repeatedly and then ranges of equalization on each regions. The experimental results performed with image of various contrasts show that the proposed algorithm is better than the global histogram equalization.

• The Journal of the Acoustical Society of Korea
• /
• v.30 no.6
• /
• pp.308-314
• /
• 2011

This paper presents techniques for estimating directions of multiple sound sources ranging from $0^{\circ}$ to $360^{\circ}$ using circular probability distributions having a periodic property. Phase differences containing direction information of sources can be modeled as mixtures of multiple probability distributions and source directions can be estimated by maximizing log-likelihood functions. Although the von Mises distribution is widely used for analyzing this kind of periodic data, we define a new class of circular probability distributions from Gaussian and Laplacian distributions by adopting a modulo operation to have $2{\pi}$-periodicity. Direction estimation with these circular probability distributions is done by implementing corresponding EM (Expectation-Maximization) algorithms. Simulation results in various reverberant environments confirm that Laplacian distribution provides better performance than von Mises and Gaussian distributions.