강수량의 계절 예측을 위한 베이지안 앙상블 MOS방법의 비교연구

조성일,이상인 한국데이터정보과학회 2019 한국데이터정보과학회지 Vol.30 No.2

본 논문은 기후예측 (climate forecasts)에 있어서 주로 사용되는 통계적 후처리 (statistical post-processing)방법을 검토한다. 특히 베이지안 통계학 (Bayesian statistics)을 이용한 기후예측에서 주로 사용되는 베이지안 선형 회귀모형 (Bayesian linear regression)과 베이지안 모형 평균화 (Bayesian model averaging) 두 가지 앙상블 MOS (ensemble model output statistics)방법을 설명하고 디리크레 과정 사전분포 (Dirichlet process prior)를 이용한 비모수 (nonparametric) 베이지안 접근법을 살펴본다. 세 가지 베이지안 앙상블 방법을 바탕으로 사후분포를 유도하고 마코프 체인 몬테 카를로 (Markov chain Monte Carlo) 방법을 통해 사후추론을 실시한다. 한국 지역의 강수량 자료로 부터 leave-one-out 교차타당성 (cross-validation) 방법을 이용하여 모형간의 성능을 비교한다. 모의 실험의 결과 베이지안 통계적 후처리 방법이 일반 순환모형보다 우수한 성능을 보이는 것을 확인하였다. This paper studies statistical post-processing methods for climate forecasts. In particular, we describe Bayesian linear regression model and Bayesian model averaging method which are the most popular methods, and explain a Bayesian nonparametric model using a Dirichlet process prior as an alternative of ensemble model output statistics. Based on three Bayesian ensemble model output statistics methods, the posterior distributions are derived and the posterior inferences are performed via Markov chain Monte Carlo methods. We compare three Bayesian ensemble model output statistics methods using leave-one-out cross-validation with precipitation data over Korean peninsula. The results show that the Bayesian ensemble model output statistics methods perform better than the general circulation model.

Probabilistic shear strength models for reinforced concrete beams without shear reinforcement

송준호,김강수,Won-Hee Kang,Sungmoon Jung 국제구조공학회 2010 Structural Engineering and Mechanics, An Int'l Jou Vol.34 No.1

In order to predict the shear strengths of reinforced concrete beams, many deterministic models have been developed based on rules of mechanics and on experimental test results. While the constant and variable angle truss models are known to provide reliable bases and to give reasonable predictions for the shear strengths of members with shear reinforcement, in the case of members without shear reinforcement, even advanced models with complicated procedures may show lack of accuracy or lead to fairly different predictions from other similar models. For this reason, many research efforts have been made for more accurate predictions, which resulted in important recent publications. This paper develops probabilistic shear strength models for reinforced concrete beams without shear reinforcement based on deterministic shear strength models, understanding of shear transfer mechanisms and influential parameters, and experimental test results reported in the literature. Using a Bayesian parameter estimation method, the biases of base deterministic models are identified as algebraic functions of input parameters and the errors of the developed models remaining after the bias-correction are quantified in a stochastic manner. The proposed probabilistic models predict the shear strengths with improved accuracy and help incorporate the model uncertainties into vulnerability estimations and risk-quantified designs.

Bayesian evolutionary hypernetworks for interpretable learning from high-dimensional data

Kim, Soo-Jin,Ha, Jung-Woo,Kim, Heebal,Zhang, Byoung-Tak Elsevier 2019 Applied soft computing Vol.81 No.-

<P><B>Abstract</B></P> <P>Higher-order representation is suitable for the complicated relationships among many factors. However, existing higher-order classification models have difficulties in learning from high-dimensional data due to their large combinatorial hypothesis spaces. The interpretability of models is also significant for causality analysis. Here we propose a Bayesian evolutionary method to learn a higher-order graphical model for high-dimensional data, called Bayesian evolutionary hypernetwork (BEHN). Our method represents the combinatorial feature space using a generalized graph, hypernetwork. A hypernetwork contains a large population of hyperedges encoding higher-order relationships among feature variables, and is optimized by an evolutionary algorithm formulated as sequential Bayesian sampling. This Bayesian evolutionary approach allows for probabilistic search through the higher-order feature space while satisfying soft constraints defined by the priors. We show that two information-theoretic and complexity-related priors are effective to balance model accuracy and parsimony. Also, BEHN provides interpretable representations to investigate feature interactions. Using two benchmarking and three real-world datasets we demonstrate that BEHN outperforms baseline classification models while tackling large-scale data of dimensionality up to <I>O</I>( 1 <SUP> 0 4 </SUP> ). We also analyze the stability and the scalability of the proposed method with respect to accuracy, computational cost, and the interpretability of the model structures.</P> <P><B>Highlights</B></P> <P> <UL> <LI> We propose a new hypergraph-based model for high-dimensional data (BEHN). </LI> <LI> We describe Bayesian evolutionary approach for learning our hypergraph model. </LI> <LI> BEHN employs two information-theoretic and complexity-regulating priors. </LI> <LI> The evolutionary learning of BEHN is formulated as a sequential Bayesian sampling. </LI> <LI> BEHN provides an interpretable result based on flexible hypergraph structures. </LI> <LI> BEHN is evaluated on real-world datasets including tens of thousands of variables. </LI> </UL> </P> <P><B>Graphical abstract</B></P> <P>[DISPLAY OMISSION]</P>

Bayesian degradation modeling for reliability prediction of organic light-emitting diodes

Bae, S.J.,Yuan, T.,Kim, S.j. Elsevier 2016 Journal of computational science Vol.17 No.1

<P>Simpler degradation models are generally preferred to simplify analytical procedure of failure-time estimation which follows the degradation modeling. However, the luminosity degradation of organic light-emitting diode (OLED) tends to exhibit an initial unstable period followed by stable and more gradual degradation. The degradation mechanisms of OLED luminosity are illustrated via a stochastic two-compartment model. Conjoining the data with prior information accumulated from field testing, we propose two hierarchical Bayesian models to characterize the nonlinear degradation path of OLED: Bayesian change-point regression model and Bayesian bi-exponential model. The hierarchical Bayesian models effectively fit the nonlinear degradation paths of OLEDs. Analytical results of OLED degradation indicate that reliability estimation from the hierarchical Bayesian models can be substantially improved over the log-linear model which has been widely accepted as a degradation model of light displays. (C) 2016 Published by Elsevier B.V.</P>

클러스터 엔진 시스템의 신뢰도 분석을 위한 베이지안 계층 모델링

유승우(Seung-Woo Yoo),신명호(Myoung Ho Shin) 대한기계학회 2021 대한기계학회 춘추학술대회 Vol.2021 No.4

본 논문에서는 가용한 사전 정보를 계층화 하여 표현하고, 지상연소시험 데이터를 업데이트하는 베이지안 계층 모형을 활용하여 액체로켓엔진의 신뢰도를 추정하였다. 모수 추정을 위하여 다양한 유형과 종류의 정보를 통합하여 활용하는 베이지안 계층 모델링을 위하여 사전분포 또는 관측 데이터 정보를 계층 구조로 구성하고, 액체로켓엔진의 고장 유형에 따라 2 가지의 베이지안 계층 모델을 병합하였다. 액체로켓엔진의 점화, 시동 및 천이구간에서 발생하는 고온 열충격, 점화 충격, 진동 등에 의한 부하증가형 고장은 이산형 분포에 해당되므로 베이지안 계층 이항 모델을 이용하고, 연소시간이 증가함에 따라 강도가 저하되거나 마모, 삭마, 열변형, 연소생성물의 침착, 피로 등의 고장 모드로 나타나는 강도 저하형 고장은 베이지안 계층 와이블 모델로 나타냈다. 이항 모델에서 개별 엔진의 임무 성공률은 이항분포로 가정하고, 이에 대한 사전분포는 공액사전분포인 베타분포, 초사전분포는 감마함수로 모델링하였다. 와이블 모델에서 고장발생 시간은 복합 시스템의 고장 분포에 근사한 와이블 분포로 모사하고, 고장이 발생하지 않은 시험에서의 사전 계획시간은 우측 중도 절단 데이터로 대입하였다. 클러스터 엔진 시스템은 우주발사체의 추진 및 제어를 위해 복수의 엔진을 결합하여 추력 성능을 얻는 시스템으로, 동일한 형상의 엔진을 다발 형태로 결합하고 개별적인 시동 및 제어시스템을 적용하는 반면 추진제 공급 계통은 공통으로 활용한다는 특징이 있다. 최종적으로 클러스터 엔진 시스템의 신뢰도를 추정하기 위해 개별 엔진의 시험 데이터를 이용하여 추정한 신뢰도와 클러스터 엔진 시스템의 시험 데이터로부터 추정한신뢰도를 함께 고려하였으며, 베이지안 통합 기법을 이용하여 신뢰도를 추정한 결과를 제시하였다. The reliability of liquid rocket engine is estimated by using the Bayesian hierarchical model that expresses available prior information hierarchically and updates based on the ground combustion test data. For Bayesian hierarchical modeling that integrates and utilizes various types and types of information for parameter estimation, prior distribution or observation data information is organized in a hierarchical structure, and two Bayesian hierarchical models are merged according to the failure type of the liquid rocket engine. Load-increasing failures caused by high-temperature thermal impact, ignition shock, and vibration occurring in the ignition, start and transition section of a liquid rocket engine are discrete distribution, we use the Bayesian hierarchical binomial model, and the strength decreases as the combustion time increases Intensity-degraded failures, which appear in failure modes such as wear, ablation, thermal deformation, deposition of combustion products, and fatigue, are represented by Bayesian hierarchical Weibull models. In the binomial model, the mission success rate of individual engines is assumed to be a binomial distribution, and the prior distributions are modeled as conjugated distributions of beta distribution, and the hyperprior distributions are modeled as gamma functions. In the Weibull model, the failure occurrence time was simulated as a Weibull distribution that approximates the failure distribution of the complex system, and the pre-planned time in the test where no failure occurred was substituted with the right censored data. The cluster engine system is a system that obtains thrust performance by combining multiple engines for propulsion and control of the space launch vehicle. The engines of the same shape are combined in bundle and individual engine start and control systems are applied, while the propellant supply system is common. Finally, to estimate the reliability of the cluster engine system, the reliability estimated from the individual engine test data and the combustion test or flight test results at the level of the clustered engine were considered together, and the estimated Bayesian reliability were presented.

다중 강우유출자료를 이용한 Clark 단위도의 Bayesian 매개변수 추정

김진영,권덕순,배덕효,권현한 한국수자원학회 2020 한국수자원학회논문집 Vol.53 No.5

The main objective of this study is to provide a robust model for estimating parameters of the Clark unit hydrograph (UH) using the observed rainfall-runoff data in the Soyangang dam basin. In general, HEC-1 and HEC-HMS models, developed by the Hydrologic Engineering Center, have been widely used to optimize the parameters in Korea. However, these models are heavily reliant on the objective function and sample size during the optimization process. Moreover, the optimization process is carried out on the basis of single rainfall-runoff data, and the process is repeated for other events. Their averaged values over different parameter sets are usually used for practical purposes, leading to difficulties in the accurate simulation of discharge. In this sense, this paper proposed a hierarchical Bayesian model for estimating parameters of the Clark UH model. The proposed model clearly showed better performance in terms of Bayesian inference criterion (BIC). Furthermore, the result of this study reveals that the proposed model can also be applied to different hydrologic fields such as dam design and design flood estimation, including parameter estimation for the probable maximum flood (PMF). 본 연구에서는 소양강댐 유역에서의 실측 단일사상 강우-유출 자료를 활용하여 Clark 단위도 방법의 매개변수를 최적화 하였으며, 그 결과를 제시하였다. 일반적으로 국내에서는 유역특성인자 최적화 분석시 미육군공병단의 HEC-1, HEC-HMS 등의 모형을 사용하고 있다. 그러나 해당 모형의 경우 유출수문곡선의 형상, 크기 등의 재현에만 초점이 맞춰져 있으며, 산정된 매개변수들의 평균을 사용하고 있어 실제 강우-유출 관계를 묘사하는데 어려움이 존재하고 있다. 이러한 점에서 본 연구에서는 기존 Clark 합성단위도법과 계층적 Bayesian 기법을 결합하여 수집된 강우-유출 자료를 동시에 활용하여 매개변수를 산정할 수 있는 모형을 개발하였다. 본 연구에서 개발된 모형을 적용한 결과 개별 단일사상 기반의 최적화 기법에 비해 다중 강우-유출 자료를 Pooling하여 매개변수를 산정하는 계층적 Bayesian 모형에서 BIC 결과 및 다수의 통계적 지표를 통해 모형의 우수성을 확인할 수 있었다. 더불어 홍수량에 따른 유역특성인자 매개변수 반응에 대한 관계규명을 기반으로 향후 댐 설계 또는 PMF 산정시 본 연구의 결과가 활용이 가능할 것으로 판단된다.

Probabilistic shear strength models for reinforced concrete beams without shear reinforcement

Song, Jun-Ho,Kang, Won-Hee,Kim, Kang-Su,Jung, Sung-Moon Techno-Press 2010 Structural Engineering and Mechanics, An Int'l Jou Vol.34 No.1

In order to predict the shear strengths of reinforced concrete beams, many deterministic models have been developed based on rules of mechanics and on experimental test results. While the constant and variable angle truss models are known to provide reliable bases and to give reasonable predictions for the shear strengths of members with shear reinforcement, in the case of members without shear reinforcement, even advanced models with complicated procedures may show lack of accuracy or lead to fairly different predictions from other similar models. For this reason, many research efforts have been made for more accurate predictions, which resulted in important recent publications. This paper develops probabilistic shear strength models for reinforced concrete beams without shear reinforcement based on deterministic shear strength models, understanding of shear transfer mechanisms and influential parameters, and experimental test results reported in the literature. Using a Bayesian parameter estimation method, the biases of base deterministic models are identified as algebraic functions of input parameters and the errors of the developed models remaining after the bias-correction are quantified in a stochastic manner. The proposed probabilistic models predict the shear strengths with improved accuracy and help incorporate the model uncertainties into vulnerability estimations and risk-quantified designs.

Bayesian ballast damage detection utilizing a modified evolutionary algorithm

Hu, Qin,Lam, Heung Fai,Zhu, Hong Ping,Alabi, Stephen Adeyemi Techno-Press 2018 Smart Structures and Systems, An International Jou Vol.21 No.4

This paper reports the development of a theoretically rigorous method for permanent way engineers to assess the condition of railway ballast under a concrete sleeper with the potential to be extended to a smart system for long-term health monitoring of railway ballast. Owing to the uncertainties induced by the problems of modeling error and measurement noise, the Bayesian approach was followed in the development. After the selection of the most plausible model class for describing the damage status of the rail-sleeper-ballast system, Bayesian model updating is adopted to calculate the posterior PDF of the ballast stiffness at various regions under the sleeper. An obvious drop in ballast stiffness at a region under the sleeper is an evidence of ballast damage. In model updating, the model that can minimize the discrepancy between the measured and model-predicted modal parameters can be considered as the most probable model for calculating the posterior PDF under the Bayesian framework. To address the problems of non-uniqueness and local minima in the model updating process, a two-stage hybrid optimization method was developed. The modified evolutionary algorithm was developed in the first stage to identify the important regions in the parameter space and resulting in a set of initial trials for deterministic optimization to locate all most probable models in the second stage. The proposed methodology was numerically and experimentally verified. Using the identified model, a series of comprehensive numerical case studies was carried out to investigate the effects of data quantity and quality on the results of ballast damage detection. Difficulties to be overcome before the proposed method can be extended to a long-term ballast monitoring system are discussed in the conclusion.

시계열 범주형 자료를 위한 동적 베이지안 모형 비교

최보승 한국자료분석학회 2012 Journal of the Korean Data Analysis Society Vol.14 No.6

We considered a dynamic Bayesian model for model fitting and forecasting for contingency table data according to time tendency. We introduced and utilized the dynamic Bayesian model of Park et al. (2012) and performed various simulation studies to measure the performance of dynamic Bayesian model whether the model can capture the trend which the data may have. According to the simulation study results, the proposed model shows better performance based on MSE and biases for all four hypothesis model. However, the model that have stochastic and deterministic trend simultaneously shows better performance based on Bayesian model selection. The proposed model applied to the pre election survey for US Governor race of Ohio, November 1998. 본 연구는 시간의 흐름에 따라 관찰된 분할표 자료에 대한 모형 적합 및 예측을 위한 통계적 모형으로 동적 베이지안 모형을 고려하였다. Park et al.(2012)이 제안한 동적 베이지안 모형을 이용하여 관찰된 자료들이 확률적 추세를 가지고 있을 때 제안하고 있는 모형의 성능을 측정하기 위하여 다양한 모의실험을 수행하였다. 모의실험은 세 가지 측면에서 수행되었다. MSE나 bias측면에서 비교되었을 때 모든 가설모형이 적절한 결과를 보여 주었고 이에 반하여 Bayesian 모형 선택 측면에서 보았을 때는 확률적 추세와 결정적 추세를 모두 가지고 있는 모형에서 좋은 성능을 보여 주었다. 제안된 모형은 1998년 미국 오하이오주 주지사 선거를 앞두고 시행한 사전조사 결과에 적용하여 모형 적합 및 예측을 수행하였다.

Structural health monitoring of Canton Tower using Bayesian framework

Sin-Chi Kuok,Ka-Veng Yuen 국제구조공학회 2012 Smart Structures and Systems, An International Jou Vol.10 No.4

This paper reports the structural health monitoring benchmark study results for the Canton Tower using Bayesian methods. In this study, output-only modal identification and finite element model updating are considered using a given set of structural acceleration measurements and the corresponding ambient conditions of 24 hours. In the first stage, the Bayesian spectral density approach is used for output-only modal identification with the acceleration time histories as the excitation to the tower is unknown. The modal parameters and the associated uncertainty can be estimated through Bayesian inference. Uncertainty quantification is important for determination of statistically significant change of the modal parameters and for weighting assignment in the subsequent stage of model updating. In the second stage, a Bayesian model updating approach is utilized to update the finite element model of the tower. The uncertain stiffness parameters can be obtained by minimizing an objective function that is a weighted sum of the square of the differences (residuals) between the identified modal parameters and the corresponding values of the model. The weightings distinguish the contribution of different residuals with different uncertain levels. They are obtained using the Bayesian spectral density approach in the first stage. Again, uncertainty of the stiffness parameters can be quantified with Bayesian inference. Finally, this Bayesian framework is applied to the 24- hour field measurements to investigate the variation of the modal and stiffness parameters under changing ambient conditions. Results show that the Bayesian framework successfully achieves the goal of the first task of this benchmark study.