• 한국정보과학회 언어공학연구회:학술대회논문집(한글 및 한국어 정보처리)
• /
• 한국정보과학회언어공학연구회 2021년도 제33회 한글 및 한국어 정보처리 학술대회
• /
• pp.287-292
• /
• 2021

좋은 자연어 이해 시스템은 인간과 같이 텍스트에서 단순히 단어나 문장의 형태를 인식하는 것 뿐만 아니라 실제로 그 글이 의미하는 바를 정확하게 추론할 수 있어야 한다. 이 논문에서 우리는 뉴스 헤드라인으로 뉴스의 토픽을 분류하는 open benchmark인 KLUE(Korean Language Understanding Evaluation)에 대하여 기존에 비교 실험이 진행되지 않은 시중에 공개된 다양한 한국어 라지스케일 모델들의 성능을 비교하고 결과에 대한 원인을 실증적으로 분석하려고 한다. KoBERT, KoBART, KoELECTRA, 그리고 KcELECTRA 총 네가지 베이스라인 모델들을 주어진 뉴스 헤드라인을 일곱가지 클래스로 분류하는 KLUE-TC benchmark에 대해 실험한 결과 KoBERT가 86.7 accuracy로 가장 좋은 성능을 보여주었다.

• 제어로봇시스템학회논문지
• /
• 제7권11호
• /
• pp.896-903
• /
• 2001

This study investigates the use of mixed $H_2/H_{$\infty}$ and $\mu$-synthesis to construct a robust controller for the benchmark problem. The model treated in the problem is a coupled three-inertial system that reflects the dynamics of mechanical vibrations. This kind of problem requires to be satisfied the robust performance (both in the time and frequency-domain specifications). We, first, adopt the mixed $H_2/H_{$\infty}$ theory to design a feedback controller K(s). Next, $\mu$-synthesis method is applied to the overall system to make use of structured parametric uncertainty. This process permits higher levels of controller authority and reduces the conservativeness of the controller. Finally, the feedforward controller is also used to improve the transient response of the output. We confirm that all design specifications except a complementary sensitivity condition can be achieved.

• 국제초고층학회논문집
• /
• 제1권2호
• /
• pp.87-97
• /
• 2012

The indoor environments in high-rise buildings are generally well enclosed by defined boundary conditions. Here, a numerical simulation method based on the Lattice Boltzmann method (LBM), which aims to model and simulate the turbulent flow accurately in an enclosed environment, and its comparison with traditional computational fluid dynamics (CFD) results, are presented in this paper. CFD has become a powerful tool for predicting and evaluating enclosed airflows with the rapid advance in computer capacity and speed, and various types of CFD turbulence modeling and its application and validation have been reported. The LBM is a relatively new method; it involves solving of the discrete Boltzmann equation to simulate the fluid flow with a collision model instead of solving Navier-Stokes equations. In this study, the LBM-based scheme of flow pattern and particle dispersion analyses are validated using the benchmark test case of two- and three-dimensional and isothermal conditions (IEA/Annex 20 case); the prediction accuracy and advantages are also discussed by comparison with the results of CFD.

• Smart Structures and Systems
• /
• 제30권6호
• /
• pp.557-569
• /
• 2022

Vibration-based structural health monitoring (SHM) is crucial for the dynamic maintenance of civil building structures to protect property security and the lives of the public. Analyzing these vibrations with modern artificial intelligence and deep learning (DL) methods is a new trend. This paper proposed an unsupervised deep learning method based on a convolutional autoencoder (CAE), which can overcome the limitations of conventional supervised deep learning. With the convolutional core applied to the DL network, the method can extract features self-adaptively and efficiently. The effectiveness of the method in detecting damage is then tested using a benchmark model. Thereafter, this method is used to detect damage and instant disaster events in a rubber bearing-isolated gymnasium structure. The results indicate that the method enables the CAE network to learn the intact vibrations, so as to distinguish between different damage states of the benchmark model, and the outcome meets the high-dimensional data distribution characteristics visualized by the t-SNE method. Besides, the CAE-based network trained with daily vibrations of the isolating layer in the gymnasium can precisely recover newly collected vibration and detect the occurrence of the ground motion. The proposed method is effective at identifying nonlinear variations in the dynamic responses and has the potential to be used for structural condition assessment and safety warning.

• Smart Structures and Systems
• /
• 제31권2호
• /
• pp.199-212
• /
• 2023

Nowadays fuzzy logic in control applications is a well-recognized alternative, and this is thanks to its inherent advantages. Generalized type-2 fuzzy sets allow for a third dimension to capture higher order uncertainty and therefore offer a very powerful model for uncertainty handling in real world applications. With the recent advances that allowed the performance of general type-2 fuzzy logic controllers to increase, it is now expected to see the widespread of type-2 fuzzy logic controllers to many challenging applications in particular in problems of structural control, that is the case study in this paper. It should be highlighted that this is the first application of general type-2 fuzzy approach in civil structures. In the following, general type-2 fuzzy logic controller (GT2FLC) will be used for active control of a 9-story nonlinear benchmark building. The design of type-1 and interval type-2 fuzzy logic controllers is also considered for the purpose of comparison with the GT2FLC. The performance of the controller is validated through the computer simulation on MATLAB. It is demonstrated that extra design degrees of freedom achieved by GT2FLC, allow a greater potential to better model and handle the uncertainties involved in the nature of earthquakes and control systems. GT2FLC outperforms successfully a control system that uses T1 and IT2 FLCs.

• Nuclear Engineering and Technology
• /
• 제55권2호
• /
• pp.756-768
• /
• 2023

Neutron transport calculations are extremely challenging due to the high computational cost of large and complex problems. A multilevel octree grid algorithm (MLTG) of discrete ordinates method was developed to improve the modeling accuracy and simulation efficiency on 3-D Cartesian grids. The Balakovo-3 VVER-1000 neutron dosimetry benchmark is calculated to verify and validate this numerical technique. A simplified S₂ synthetic acceleration is used in the MLTG calculation method to improve the convergence of the source iterations. For the triangularly arranged fuel pins, we adopt a source projection algorithm to generate pin-by-pin source distributions of hexagonal assemblies. MLTG provides accurate geometric modeling and flexible fixed source description at a lower cost than traditional Cartesian grids. The total number of meshes is reduced to 1.9 million from the initial 9.5 million for the Balakovo-3 model. The numerical comparisons show that the MLTG results are in satisfactory agreement with the conventional S_N method and experimental data, within the root-mean-square errors of about 4% and 10%, respectively. Compared to uniform fine meshing, approximately 70% of the computational cost can be saved using the MLTG algorithm for the Balakovo-3 computational model.

• 방사선산업학회지
• /
• 제17권4호
• /
• pp.333-344
• /
• 2023

To validate the numerical model used in the study of deep disposal of spent nuclear fuel, we selected benchmark cases and performed numerical model validation. We selected the DECOVALEX-THMC Task D_THM1 FEBEX Type benchmark case, which was conducted from 2003 to 2007. We analyzed the thermal-hydraulic (TH) behavior using the finite element program CODE_BRIGHT and verified the results against previous studies. The temperature results were similar to the results of DECOVALEX-THMC Task D. The saturation results showed a similar trend to the results of DECOVALEX-THMC Task D, but the time to reach full saturation was different.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2000년도 추계종합학술대회 논문집(4)
• /
• pp.85-88
• /
• 2000

We present a new approach to analyzing the dynamic false contour noise of AC plasma display panels (PDP), which is known to degrade the image quality severely. Compared with the existing methods that consider only the amount of light emission from PDP during 1 field time, the proposed approach uses the impulse response model of the human vision system and estimates how the human beings actually feel as the function of time. Experimental results using various benchmark sub-field scan algorithms are included.

• 응용통계연구
• /
• 제35권2호
• /
• pp.229-250
• /
• 2022

포트폴리오 최적화 이론의 초석인 Markowitz의 평균-분산 포트폴리오 모형 (1952)이 발표된 이후로 많은 분야에서 포트폴리오 최적화에 대한 다양한 연구가 진행되었다. 기존의 평균-분산 포트폴리오 모형은 주로 목적함수나 제약식에 비선형 볼록 형태를 포함한다. 이를 Dantzig의 선형계획법을 적용하여 선형으로 변환시켜 알고리즘 계산 시간을 효율적으로 감소시켰다. 또한 시계열 데이터 특성을 반영하여 시간에 따른 가중치를 고려하는 가우시안 커널 가중치 공분산을 제안하였다. 여기에 일정 부분은 벤치마크에 투자하고 나머지는 포트폴리오 최적화 모형으로 제안된 자산들에 투자하는 퍼터베이션 방법을 적용하여 평균 수익률과 위험도를 목적에 맞게 조절하도록 하였다. 또한, 본 논문에서는 안정적이면서도 적은 자산을 보유하게 포트폴리오를 구성하여 관리비용(management costs)과 거래비용(transaction costs)를 낮출 수 있는 Dantzig-type 퍼터베이션 포트폴리오 모형을 제안하였다. 제안된 모형의 성능은 5개의 실제 데이터 세트로 벤치마크 포트폴리오와 비교 분석하여 평가하였다. 최종적으로 제안한 최적화 모형은 벤치마크보다 높은 기대수익률이나 낮은 위험도를 갖는 포트폴리오를 구성하여 퍼터베이션 목적을 만족하며, 투자한 자산의 수와 시간에 따른 자산 구성 변화를 일정 수준 이하로 조절하는 희소하며 안정적인 결과를 얻었다.

• Nuclear Engineering and Technology
• /
• 제55권10호
• /
• pp.3874-3897
• /
• 2023

A high-fidelity computational fluid dynamics (CFD) analysis was performed using the Large Eddy Simulation (LES) model for the lower plenum of the High-Temperature Test Facility (HTTF), a ¼ scale test facility of the modular high temperature gas-cooled reactor (MHTGR) managed by Oregon State University. In most next-generation nuclear reactors, thermal stress due to thermal striping is one of the risks to be curiously considered. This is also true for HTGRs, especially since the exhaust helium gas temperature is high. In order to evaluate these risks and performance, organizations in the United States led by the OECD NEA are conducting a thermal hydraulic code benchmark for HTGR, and the test facility used for this benchmark is HTTF. HTTF can perform experiments in both normal and accident situations and provide high-quality experimental data. However, it is difficult to provide sufficient data for benchmarking through experiments, and there is a problem with the reliability of CFD analysis results based on Reynolds-averaged Navier-Stokes to analyze thermal hydraulic behavior without verification. To solve this problem, high-fidelity 3-D CFD analysis was performed using the LES model for HTTF. It was also verified that the LES model can properly simulate this jet mixing phenomenon via a unit cell test that provides experimental information. As a result of CFD analysis, the lower the dependency of the sub-grid scale model, the closer to the actual analysis result. In the case of unit cell test CFD analysis and HTTF CFD analysis, the volume-averaged sub-grid scale model dependency was calculated to be 13.0% and 9.16%, respectively. As a result of HTTF analysis, quantitative data of the fluid inside the HTTF lower plenum was provided in this paper. As a result of qualitative analysis, the temperature was highest at the center of the lower plenum, while the temperature fluctuation was highest near the edge of the lower plenum wall. The power spectral density of temperature was analyzed via fast Fourier transform (FFT) for specific points on the center and side of the lower plenum. FFT results did not reveal specific frequency-dominant temperature fluctuations in the center part. It was confirmed that the temperature power spectral density (PSD) at the top increased from the center to the wake. The vortex was visualized using the well-known scalar Q-criterion, and as a result, the closer to the outlet duct, the greater the influence of the mainstream, so that the inflow jet vortex was dissipated and mixed at the top of the lower plenum. Additionally, FFT analysis was performed on the support structure near the corner of the lower plenum with large temperature fluctuations, and as a result, it was confirmed that the temperature fluctuation of the flow did not have a significant effect near the corner wall. In addition, the vortices generated from the lower plenum to the outlet duct were identified in this paper. It is considered that the quantitative and qualitative results presented in this paper will serve as reference data for the benchmark.