• Nuclear Engineering and Technology
• /
• 제56권6호
• /
• pp.2266-2273
• /
• 2024

A machine learning platform is proposed for the diagnosis of a severe accident progression in a nuclear power plant. To predict the key parameters for accident management including lost signals, a long short term memory (LSTM) network is proposed, where multiple accident scenarios are used for training. Training and test data were produced by MELCOR simulation of the Fukushima Daiichi Nuclear Power Plant (FDNPP) accident at unit 3. Feature variables were selected among plant parameters, where the importance ranking was determined by a recursive feature elimination technique using RandomForestRegressor. To answer the question of whether a reduced order ML model could predict the complex transient response, we performed a systematic sensitivity study for the choices of target variables, the combination of training and test data, the number of feature variables, and the number of neurons to evaluate the performance of the proposed ML platform. The number of sensitivity cases was chosen to guarantee a 95 % tolerance limit with a 95 % confidence level based on Wilks' formula to quantify the uncertainty of predictions. The results of investigations indicate that the proposed ML platform consistently predicts the target variable. The median and mean predictions were close to the true value.

• 한국정보처리학회논문지
• /
• 제5권11호
• /
• pp.2823-2834
• /
• 1998

본 논문에서는 병렬 프로그램을 구성하는 $2^n$개의 타스크 모듈들을 n-차원 하이퍼큐브 다중 컴퓨터에 전체 통신 비용이 최소가 되도록 일대일 매핑하는 문제를 다룬다. 하이퍼큐브에서 최적 매핑을 구한 것은 NP-complete문제이다. 본 논문에서는 먼저 하이퍼큐브 다중 컴퓨터에서의 매핑 문제를 그래프 상에서의 최대 컷세트 집합을 구하는 문제로 변환시키는 그래프 변형 기법을 제안한다. 이러한 그래프 변형 기법을 사용하여 기존의 그래프 이분할 방법을 변형된 그래프 상에 반복 적용함으로써 하이퍼큐브에 타스크 모듈들을 효율적으로 일대일 매핑하는 반복 매핑 알고리즘을 제안한다. 여러가지 타스크그래프 상에서의 실험을 통해, 제안된 반복 매핑 알고리즘이 기존의 greedy나 recursive 매핑 알고리즘들 보다 성능이 우수함을 보인다. 특히 제안된 알고리즘은 하이퍼큐브-isomorphic, 메쉬등과 같은 정형 그래프 상에서 성능이 우수하며 거의 모든 정형 그래프에서 최적 매핑을 찾음을 보인다.

• 한국전자파학회논문지
• /
• 제27권3호
• /
• pp.317-325
• /
• 2016

본 논문은 추적-후-탐지 처리(TBD: Track Before Detect)를 위한 레이더용 파티클 필터(Particle filter)에 대해서 기술한다. TBD 기법은 강한 클러터 환경, 작은 RCS 타겟 및 스텔스 타겟 등으로 인해 타겟 탐지가 어려운 경우(낮은 SNR)에 적용하는 기술이다. 특히 파티클 필터는 재귀적 TBD(Recursive TBD) 알고리즘 구현에 적합하고, 비선형 모델을 가우시안 선형 모델로 근사화해서 추정하는 칼만 필터 대비, 상대적으로 개선된 정확도를 갖는다. 본 논문에서는 다수의 관측값(클러터 포함)들이 동시에 수신될 때, 신호강도-거리-도플러 정보를 활용하여 파티클 필터 가중치를 직접 계산 및 갱신하는 방식을 제안한다. 성능 분석을 위해 가상의 레이더 시뮬레이션 사니리오를 설정하고, 제안하는 파티클 필터를 적용하여 추적 필터의 추정오차를 분석한다.

• KNOM Review
• /
• 제24권2호
• /
• pp.13-23
• /
• 2021

최근 IoT 기기들의 활성화에 의해 네트워크가 복잡해짐에 따라, 네트워크의 혼잡을 예측하고 미리 대비하기 위해 단기 트래픽 예측을 넘어 장기 트래픽 예측 연구가 활성화되고 있다. 단기 트래픽 예측 결과를 입력으로 재사용하는 재귀 전략은 멀티 스텝 트래픽 예측으로 확장되었지만, 재귀 단계가 진행될수록 오류가 축적되어 예측 성능 저하를 일으킨다. 이 논문에서는 다중 출력 전략을 사용한 LSTM 기반 멀티스텝 트래픽 예측 기법을 소개하고그 성능을 평가한다. 실제 DNS 요청 트래픽을 기반으로 실험한 결과, 제안된 LSTM기반 다중출력 전략 기법은 재귀 전략 기법에 비해 비정상성 트래픽에 대한 트래픽 예측 성능의 MAPE를 약 6% 줄일 수 있음을 확인하였다.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2004년도 ICCAS
• /
• pp.1694-1700
• /
• 2004

In this paper, we propose the AWLS/MFT (Adaptive Weighed Least Squares/ Modulation Function Technique) devised by A. E. Pearson et al. for the transfer function estimation of a modal system and investigate the performance of several algorithms, the Gram matrix method, a Luenberger Observer (LO), Least Squares (LS), and Recursive Least Squares (RLS), for the estimation of initial conditions. With the benefit of the Modulation Function Technique (MFT), we can separate the estimation problem into two phases: the transfer function parameters are estimated in the first phase, and the initial conditions are estimated in the second phase. The LO method produces excellent IC estimates in the noise free case, but the other three methods show better performance in the noisy case. Finally, we compared our result with the Prony based method. In the noisy case, the AWLS and one of the three methods - Gram matrix, LS, and RLS- show better performance in the output Signal to Error Ratio (SER) aspect than the Prony based method under the same simulation conditions.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 1998년도 추계종합학술대회 논문집
• /
• pp.1081-1084
• /
• 1998

In this paper, we propose the bus coding technique for low power consumption. For CMOS circuit most power is dissipated as dynamic power for charging and discharging node capacitances.Though the I/O and bus are likely to have the very large capacitances associated with them and dissipate much of the power dissipated by an IC, they have little beenthe special target for power reduction. The conventional Bus-Invert coding method can't decrease the peak power dissipation by 50% because the additional invert signal line can invoke a transition at the time when Bus-Invert coding isn't used to code original bus data. The proposed technique always constraints the Hamming distance between data transferred sequentially to be below the half of the bus width, and thus decrease the I/O peak power dissipation and the I/O average power dissipation.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2011년도 제42회 하계학술대회
• /
• pp.197-198
• /
• 2011

A technique to manage harmonic pollution at the point of common coupling (PCC) has been presented using a control chart. The recursive least-squares technique has been used to estimate the parameters of the Thevenin equivalent load model. With the estimated data from the measured voltage and current waveform at the PCC during certain period violating upper control limit, the individual contributions to the distortion of voltage waveform at an interested harmonic frequency have been calculated and shown by simple graph.

• 전기학회논문지
• /
• 제58권4호
• /
• pp.679-687
• /
• 2009

In this paper, a subband filtering and adaptive prediction technique for analyzing harmonics in power systems is presented. In this method, the filter banks are designed to decompose odd and even order harmonics separately. The adaptive prediction has been employed reduce the transient and white noise effect in time varying harmonics detecting and tracking. The frequencies and amplitudes of the decomposed harmonics are estimated by recursive algorithm. To demonstrate the performances of the developed technique, computer simulations to the signal with the time-varying frequency and THD are carried out.

• Structural Engineering and Mechanics
• /
• 제28권2호
• /
• pp.221-238
• /
• 2008

Numerical solution to buckling analysis of beams and columns are obtained by the method of differential quadrature (DQ) and harmonic differential quadrature (HDQ) for various support conditions considering the variation of flexural rigidity. The solution technique is applied to find the buckling load of fully or partially embedded columns such as piles. A simple semi- inverse method of DQ or HDQ is proposed for determining the flexural rigidities at various sections of non-prismatic column ( pile) partially and fully embedded given the buckling load, buckled shape and sub-grade reaction of the soil. The obtained results are compared with the existing solutions available from other numerical methods and analytical results. In addition, this paper also uses a recently developed technique, known as the differential transformation (DT) to determine the critical buckling load of fully or partially supported heavy prismatic piles as well as fully supported non-prismatic piles. In solving the problem, governing differential equation is converted to algebraic equations using differential transformation methods (DT) which must be solved together with applied boundary conditions. The symbolic programming package, Mathematica is ideally suitable to solve such recursive equations by considering fairly large number of terms.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2004년도 추계학술대회 학술발표 논문집 제14권 제2호
• /
• pp.515-520
• /
• 2004

In this paper, a passive telemetry RF capacitive humidity sensor system using a RLSE(Recursive Least Square Estimation) technique is proposed. To overcome these trouble problems such as a power limitation and a estimation complexity that the general passive telemetry sensor system including It chip has, the principle of inductive coupling was applied to the modeling of a passive telemetry RF capacitive humidity sensor system and its capacitance was estimatedd by the RLSE algorithm. Specially, by introducing the optimal forgetting factor, we showed that the accuracy of its estimation was improved even in the time varying system and also the convergence time was reduced.