• International Journal of Highway Engineering
• /
• v.13 no.1
• /
• pp.33-40
• /
• 2011

The resilient modulus which is presented mechanical properties of compacted subbase material is the design parameter on the Mechanistic - Empirical pavement design guide. The compaction control method on the Mechanistic - Empirical pavement design guide will be the way to confirm whether the in-situ elastic modulus measured after the compaction meets the resilient modulus which is applied the design. The resilient modulus in this study is calculated by the neural network suggested by Korea Pavement Research Program, and degree of compaction as the existing compaction control test and plate bearing capacity test(PBT) was performed to confirm whether the in-situ elastic modulus is measured. The Light Falling Weight Deflectometer(LFWD) is additionally tested for correlation analysis between each in-situ elastic modulus and resilient modulus, and is proposed correlation equation and test interval which can reduced overall testing cost. Also, the subbase compaction control procedure based on the in-situ elastic modulus is proposed using the in-situ PBT and LFWD test result.

• The Korean Journal of Applied Statistics
• /
• v.31 no.5
• /
• pp.539-554
• /
• 2018

Effective interaction between user and device is considered an important ability of IoT devices. For some applications, it is necessary to recognize human facial expressions in real time and make accurate judgments in order to respond to situations correctly. Therefore, many researches on facial image analysis have been preceded in order to construct a more accurate and faster recognition system. In this study, we constructed an automatic recognition system for facial expressions through two steps - a facial recognition step and a classification step. We compared various models with different sets of data with pixel information, landmark coordinates, Euclidean distances among landmark points, and arctangent angles. We found a fast and efficient prediction model with only 30 principal components of face landmark information. We applied several prediction models, that included linear discriminant analysis (LDA), random forests, support vector machine (SVM), and bagging; consequently, an SVM model gives the best result. The LDA model gives the second best prediction accuracy but it can fit and predict data faster than SVM and other methods. Finally, we compared our method to Microsoft Azure Emotion API and Convolution Neural Network (CNN). Our method gives a very competitive result.

• Journal of the Society of Naval Architects of Korea
• /
• v.36 no.1
• /
• pp.82-89
• /
• 1999

A designer must make a lot of decisions in a design process. The decisions may be classified into selection decisions and compromise decisions. As the results of two decisions depends on the designer's intention it is necessary that the designer's intention should be reflected in the design systematically and precisely. As the AHP(Analytic Hierarchy Process) technique analyzes and evaluates a obscure selection problem hierarchically, designer's intention can be reflected in the design systematically. Also as qualitative attributes can be rated at quantitative criterion the designer's intention can be reflected consistently. Usually an engineering problem is a coupled problem in which a designer must select one alternative from a set of alternatives and find optimal characteristics of the alternative concurrently. As considered attributes are functions of the compromise system variables and the attributes's units and orders are different each other, attribute ratings must be normalized. This paper introduces a neural network at this normalization. So the attribute ratings can reflect designer's intention and the knowledge from his(her) experience automatically.

• Nuclear Engineering and Technology
• /
• v.42 no.2
• /
• pp.219-230
• /
• 2010

In nuclear power plants (NPPs), periodic sensor calibrations are required to assure that sensors are operating correctly. By checking the sensor's operating status at every fuel outage, faulty sensors may remain undetected for periods of up to 24 months. Moreover, typically, only a few faulty sensors are found to be calibrated. For the safe operation of NPP and the reduction of unnecessary calibration, on-line instrument calibration monitoring is needed. In this study, principal component-based auto-associative support vector regression (PCSVR) using response surface methodology (RSM) is proposed for the sensor signal validation of NPPs. This paper describes the design of a PCSVR-based sensor validation system for a power generation system. RSM is employed to determine the optimal values of SVR hyperparameters and is compared to the genetic algorithm (GA). The proposed PCSVR model is confirmed with the actual plant data of Kori Nuclear Power Plant Unit 3 and is compared with the Auto-Associative support vector regression (AASVR) and the auto-associative neural network (AANN) model. The auto-sensitivity of AASVR is improved by around six times by using a PCA, resulting in good detection of sensor drift. Compared to AANN, accuracy and cross-sensitivity are better while the auto-sensitivity is almost the same. Meanwhile, the proposed RSM for the optimization of the PCSVR algorithm performs even better in terms of accuracy, auto-sensitivity, and averaged maximum error, except in averaged RMS error, and this method is much more time efficient compared to the conventional GA method.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2008.05a
• /
• pp.1447-1451
• /
• 2008

저수지 군 연계 운영을 위한 각 댐에서의 방류량을 결정하기 위해서는 대개 각 댐의 초기 저수량, 유역 상 하류 댐의 총 저수량, 수요량, 기간별 발전 목표 달성 정도, 그리고 예상되는 미래유입량 등이 추정되어야 한다. 본 연구에서는 댐 군 연계운영을 위한 일별 최적화 모형인 CoMOM(Coordinated Multi-reservoir Operating Model, 4.2)의 상위 단계의 더 큰 단위 기간에 활용될 댐 군 연계 운영 기본 가이드라인을 신경망 기법을 활용하여 도출할 수 있을 지를 실험해 보고자 한다. 이 방법은 기본적으로 CoMOM이 제시하는 일별 운영 계획의 결과가 최선의 정책일것이라는 가정에 근거하고 있다. 즉, 주어진 상황에서 일별 CoMOM이 제시하는 결과를 교사 신호로 하여 신경망 학습을 수행하고, 이 결과를 통해 규칙(Rule)을 생성하는 과정으로 요약할 수 있다. 신경망 분석은 CoMOM이 이수기 모형인 점을 고려하여 이수기만을 대상으로 실험하였으며, 단위 분석기간을 10일로 택하여 미래 10일간의 방류량을 결정하는 것을 목표로 하였다. 신경망 모형의 입력요소로는 각 댐의 초기 유효 저수량, 유역 상 하류 댐의 총 저수량, 10일간의 수요량, 그리고 향후 한달 동안의 예상 유입량을 적용하였고, 출력요소로는 CoMOM에서 제시한 방류량 결과를 사용하였다. 모형의 유효성을 검증하기 위해 한강수계의 이수기를 대상으로 과거의 유입량 자료가 재현된다고 가정하고, 모의운영을 통하여 적합성을 분석하였다. 이를 위해 매일 단위의 실제 댐 군 연계 운영의 상황을 모의할 수 있는 실시간 시뮬레이션을 적용하였으며, 신경망 모형의 운영 기준에 의해 결정된 향후 10일 동안의 총 방류량이 해당기간 동안 동일한 양으로 나누어 방류된다는 가정 하에 모의 운영하였다. 그리고 도출된 운영 결과는 최종적으로 실적과의 평균저수량, 발전량, 여수로 방류량 비교를 통해 평가하였다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2008.05a
• /
• pp.400-404
• /
• 2008

댐 군 연계운영을 위한 기존의 많은 최적화 모형은 경제성에 기반을 둔 단일 목적 함수를 가정함으로써 현실과는 동떨어진 결과를 도출하곤 하였다. 따라서 보다 현실적인 최적화 모형이 되기 위해서는 실제 운영과정을 모사할 수 있도록 적절한 초기 가중치를 부여하여 모형을 구축하고, 상충되는 목적간의 절충안으로 파레토 프런티어(Pareto-frontier)를 제시할 수 있는 다중목적 의사결정 기법이 요구된다. 본 연구의 목적은 댐 군 연계 운영을 위한 최적화 모형으로 소개된 CoMOM(Coordinated Multi-reservoir Operating Model)의 다중목적함수에 적합한 초기 가중치를 도출할 수 있는 시스템을 제안하는 것이다. 본 연구에서는 최적화 모형에 적합한 가중치를 결정함에 있어 댐의 초기저수량과 미래의 예상 유입량과 같은 수문 조건을 감안할 필요가 있음에 주목하였다. 이것은 초기저수량과 미래에 예상되는 유입량이 작을 경우 가급적 저수에 중점을 두고, 그 반대일 경우는 발전방류에 주안점을 두는 것이 바람직하다는 사실에서 추정해 볼 수 있는 가정이다. 따라서 댐의 초기 저수량 조건과 유입량 시나리오의 다양한 수문 조건별로 가장 적합한 가중치를 찾아본 후, 수문 조건을 입력요소로, 최적 가중치를 출력요소로 갖는 신경망 모형을 활용해서 수문 조건에 적합한 가중치를 예측할 수 있는 절차를 제안한다. 이 과정에서 최적 가중치를 도출하는 것이 관건이 될 수 있는데, 이를 위해 전승목 (2008)등이 제안한 DEA기반 순위결정 절차를 활용해서 최선의 파레토 최적해와 이에 대응되는 가중치를 찾아 이를 신경망 모형의 출력요소 값으로 활용하였다. 본 연구에서 제안하는 신경망 모형은 임의의 수문 상황에 대해 이에 적합한 CoMOM의 초기 가중치를 결정해 줌으로써 CoMOM과 같은 최적화 모형의 가중치 선정에 따르는 어려움을 극복하는 데 도움이 될 수 있을 것으로 기대된다.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.3
• /
• pp.27-38
• /
• 2000

This paper proposes how the recurrent neural network controller for a Khepera mobile robot with an obstacle avoiding ability can be determined by co-adaptation of the evolution and learning, The proposed co-adaptation scheme consists of two folds： a population of NN controllers are evolved by the genetic algorithm so that the degree of obstacle avoidance might be reduced through the global searching and each NN controller is trained by CRBP learning so that the running behavior is adapted to its outer environment through the local searching. Experimental results shows that the NN controller coadapted by evolution and learning outperforms its non-learning equivalent evolved by only genetic algorithm in both the ability of obstacle avoidance and the convergence speed reaching to the required running behavior.

• The Journal of Information Systems
• /
• v.28 no.4
• /
• pp.333-358
• /
• 2019

Purpose The study aims to develop a data-based decision model for private bankers when recommending hedge funds to their customers in financial institutions. Design/methodology/approach The independent variables are set in two groups. The independent variables of the first group are aggressive investors, active investors, and risk-neutral type investors. In the second group, variables considered by private bankers include customer propensity to invest, reliability, product subscription experience, professionalism, intimacy, and product understanding. A decision-making variable for a private banker is in recommending a first-rate general private fund composed of foreign and domestic FinTech products. These contain dependent variables that include target return rate(%), fund period (months), safeguard existence, underlying asset, and hedge fund name. Findings Based on the research results, there is a 94.4% accuracy in decision-making when the independent variables (customer rating, reliability, intimacy, product subscription experience, professionalism and product understanding) are used according to the following order of relevant dependent variables: step 1 on safeguard existence, step 2 on target return rate, step 3 on fund period, and step 4 on hedge fund name. Next, a 93.7% accuracy is expected when decision-making uses the following order of dependent variables: step 1 on safeguard existence, step 2 on target return rate, step 3 on underlying asset, and step 4 on fund period. In conclusion, a private banker conducts a decision making stage when recommending hedge funds to their customers. When examining a private banker's recommendations of hedge funds to a customer, independent variables influencing dependent variables are intimacy, product comprehension, and product subscription experience according to a categorical regression model and artificial neural network analysis model.

• The Journal of the Acoustical Society of Korea
• /
• v.22 no.2
• /
• pp.135-144
• /
• 2003

A method for the automatic segmentation of speech signals is described. The method is dedicated to the construction of a large database for a Text-To-Speech (TTS) synthesis system. The main issue of the work involves the refinement of an initial estimation of phone boundaries which are provided by an alignment, based on a Hidden Market Model(HMM). Multi-layer perceptron (MLP) was used as a phone boundary detector. To increase the performance of segmentation, a technique which individually trains an MLP according to phonetic transition is proposed. The optimum partitioning of the entire phonetic transition space is constructed from the standpoint of minimizing the overall deviation from hand labelling positions. With single speaker stimuli, the experimental results showed that more than 95% of all phone boundaries have a boundary deviation from the reference position smaller than 20 ms, and the refinement of the boundaries reduces the root mean square error by about 25%.

• Journal of the Korea Society of Computer and Information
• /
• v.19 no.7
• /
• pp.17-25
• /
• 2014

In the devices that leave video records, it is an important issue to distinguish whether the input image is a real object or a photo when securing an identifying image. Using a single image and sensor, which is a simple way to distinguish the target from distance measurement has many weaknesses. Thus, this paper proposes a way to distinguish a simple photo and a real object by using stereo images. It is not only measures the distance to the target, but also checks a three-dimensional effect by making the depth map of the face area. They take pictures of the photos and the real faces, and the measured value of the depth map is applied to the learning algorithm. Exactly through iterative learning to distinguish between the real faces and the photos looked for patterns. The usefulness of the proposed algorithm was verified experimentally.