• Journal of Korea Society of Industrial Information Systems
• /
• v.28 no.5
• /
• pp.15-30
• /
• 2023

Recently, generative models based on the Transformer architecture, such as ChatGPT, have been gaining significant attention. The Transformer architecture has been applied to various neural network models, including Google's BERT(Bidirectional Encoder Representations from Transformers) sentence generation model. In this paper, a method is proposed to create a text binary classification model for determining whether a comment on Korean movie review is positive or negative. To accomplish this, a pre-trained multilingual BERT sentence generation model is fine-tuned and transfer learned using a new Korean training dataset. To achieve this, a pre-trained BERT-Base model for multilingual sentence generation with 104 languages, 12 layers, 768 hidden, 12 attention heads, and 110M parameters is used. To change the pre-trained BERT-Base model into a text classification model, the input and output layers were fine-tuned, resulting in the creation of a new model with 178 million parameters. Using the fine-tuned model, with a maximum word count of 128, a batch size of 16, and 5 epochs, transfer learning is conducted with 10,000 training data and 5,000 testing data. A text sentiment binary classification model for Korean movie review with an accuracy of 0.9582, a loss of 0.1177, and an F1 score of 0.81 has been created. As a result of performing transfer learning with a dataset five times larger, a model with an accuracy of 0.9562, a loss of 0.1202, and an F1 score of 0.86 has been generated.

• Geotechnical Engineering
• /
• v.8 no.2
• /
• pp.5-20
• /
• 1992

The physical-based and lumped-parameter hydrologic groundwater flow model for predicting the rainfall-triggered rise of groundwater levels in hillside slopes is developed in this paper to assess the risk of landslides. The developed model consists of a vertical infiltration model for unsaturated zone linked to a linear storage reservoir model(LSRM) for saturated zone. The groundwater flow model has uncertain constants like soil depttL slope angle, saturated permeability, and potential evapotranspiration and four free model parameters like a, b, c, and K. The free model parameters could be estimated from known input-output records. The BARD algorithm is uses as the parameter estimation technique which is based on a linearization of the proposed model by Gauss -Newton method and Taylor series expansion. The application to examine the capacity of prediction shows that the developed model has a potential of use in forecast systems of predicting landslides and that the optimal estimate of potential 'a' in infiltration model is the most important in the global optimum analysis because small variation of it results in the large change of the objective function, the sum of squares of deviations of the observed and computed groundwater levels. 본 논문에서는 가파른 산사면에서 산사태의 발생을 예측하기 위한 수문학적 인 지하수 흐름 모델을 개발하였다. 이 모델은 물리적인 개념에 기본하였으며, Lumped-parameter를 이용하였다. 개발된 지하수 흐름 모델은 두 모델을 조합하여 구성되어 있으며, 비포화대 흐름을 위해서는 수정된 abcd 모델을, 포화대 흐름에 대해서는 시간 지체 효과를 고려할 수 있는 선형 저수지 모델을 이용하였다. 지하수 흐름 모델은 토층의 두께, 산사면의 경사각, 포화투수계수, 잠재 증발산 량과 같은 불확실한 상수들과 a, b, c, 그리고 K와 같은 자유모델변수들을 가진다. 자유모델변수들은 유입-유출 자료들로부터 평가할 수 있으며, 이를 위해서 본 논문에서는 Gauss-Newton 방법을 이용한 Bard 알고리즘을 사용하였다. 서울 구로구 시흥동 산사태 발생 지역의 산사면에 대하여 개발된 모델을 적용하여 예제 해석을 수행함으로써, 지하수 흐름 모델이 산사태 발생 예측을 위하여 이용할 수 있음을 입증하였다. 또한, 매개변수분석 연구를 통하여, 변수 a값은 작은 변화에 대하여 목적함수값에 큰 변화를 일으키므로 a의 값에 대한 최적값을 구하는 것이 가장 중요한 요소라는 결론을 얻었다.

• Geotechnical Engineering
• /
• v.10 no.1
• /
• pp.47-62
• /
• 1994

It has been well known that the rainfall-triggered rise of groundwater levels is one of the most important factors resulting the instability of the hillside slopes. Thus, the prediction of porewater pressure is an essential step in the evaluation of landslide hazard. This study involves the development and verification of numerical groundwater flow model for the prediction of groundwater flow fluctuations accounting for both of unsatu나toed flow and saturated flow on steep hillside slopes. The first part of this study is to develop a nomerical groundwater flow model. The numerical technique chosen for this study is the finitro element method in combination with the finite difference method. The finite element method is used to transform the space derivatives and the finite difference method is used to discretize the time domain. The second part of this study is to estimate the unknown model parameters used in the proposed numerical model. There were three parameters to be estimated from input -output record $K_e$, $\psi_e$, b. The Maximum -A-Posteriori(MAP) optimization method is utilized for this purpose, . The developed model is applied to a site in Korea where two debris avalanches of large scale and many landslides of small scale were occurred. The results of example analysis show that the numerical groundwater flow model has a capacity of predicting the fluctuation of groundwater levels due to rainfall reasonably well.

• Economic and Environmental Geology
• /
• v.38 no.6 s.175
• /
• pp.663-673
• /
• 2005

This study investigates the relationship between the geochemical maps and the gold-silver deposit locations. Geochemical maps of 21 elements, which are published by KIGAM, locations of gold-silver deposits, and 1:1,000,000 scale geological map of Korea are utilized far this investigation. Pixel size of the basic geochemical maps is 250m and these data are resampled in 1km spacing for the statistical analyses. Relationship between the mine location and the geochemical data are investigated using bayesian statistics and decision tree algorithms. For the bayesian statistics, each geochemical maps are reclassified by percentile divisions which divides the data by 5, 25, 50, 75, 95, and $100\%$ data groups. Number of mine locations in these divisions are counted and the probabilities are calculated. Posterior probabilities of each pixel are calculated using the probability of 21 geochemical maps and the geological map. A prediction map of the mining locations is made by plotting the posterior probability. The input parameters for the decision tree construction are 21 geochemical elements and lithology, and the output parameters are 5 types of mines (Ag/Au, Cu, Fe, Pb/Zn, W) and absence of the mine. The locations for the absence of the mine are selected by resampling the overall area by 1 km spacing and eliminating my resampled points, which is in 750m distance from mine locations. A prediction map of each mine area is produced by applying the decision tree to every pixels. The prediction by Bayesian method is slightly better than the decision tree. However both prediction maps show reasonable match with the input mine locations. We interpret that such match indicate the rules produced by both methods are reasonable and therefore the geochemical data has strong relations with the mine locations. This implies that the geochemical rules could be used as background values oi mine locations, therefore could be used for evaluation of mine contamination. Bayesian statistics indicated that the probability of Au/Ag deposit increases as CaO, Cu, MgO, MnO, Pb and Li increases, and Zr decreases.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.24 no.5
• /
• pp.431-449
• /
• 2022

As many tunnels generally have been constructed, various experiences and techniques have been accumulated for tunnel design as well as tunnel construction. Hence, there are not a few cases that, for some usual tunnel design works, it is sufficient to perform the design by only modifying or supplementing previous similar design cases unless a tunnel has a unique structure or in geological conditions. In particular, for a tunnel blast design, it is reasonable to refer to previous similar design cases because the blast design in the stage of design is a preliminary design, considering that it is general to perform additional blast design through test blasts prior to the start of tunnel excavation. Meanwhile, entering the industry 4.0 era, artificial intelligence (AI) of which availability is surging across whole industry sector is broadly utilized to tunnel and blasting. For a drill and blast tunnel, AI is mainly applied for the estimation of blast vibration and rock mass classification, etc. however, there are few cases where it is applied to blast pattern design. Thus, this study attempts to automate tunnel blast design by means of machine learning, a branch of artificial intelligence. For this, the data related to a blast design was collected from 25 tunnel design reports for learning as well as 2 additional reports for the test, and from which 4 design parameters, i.e., rock mass class, road type and cross sectional area of upper section as well as bench section as input data as well as16 design elements, i.e., blast cut type, specific charge, the number of drill holes, and spacing and burden for each blast hole group, etc. as output. Based on this design data, three machine learning models, i.e., XGBoost, ANN, SVM, were tested and XGBoost was chosen as the best model and the results show a generally similar trend to an actual design when assumed design parameters were input. It is not enough yet to perform the whole blast design using the results from this study, however, it is planned that additional studies will be carried out to make it possible to put it to practical use after collecting more sufficient blast design data and supplementing detailed machine learning processes.

• Journal of Biomedical Engineering Research
• /
• v.41 no.2
• /
• pp.107-119
• /
• 2020

The deep neural networks (DNN) that can replicate the behavior of the human expert who recognizes the characteristics of ECG waveform have been developed and studied to analyze ECG. However, although the existing DNNs can not provide the explanations for their decisions, those trials have attempted to determine whether patients have certain diseases or not and those decisions could not be accepted because of the absence of relating theoretical basis. In addition, these DNNs required a lot of training data to obtain sufficient accuracy in spite of the difficulty in the acquisition of relating clinical data. In this study, a small-sized continuous data processing DNN (C-DNN) was suggested to determine the simple characteristics of ECG wave that were not required additional explanations about its decisions and the C-DNN can be easily trained with small training data. Although it can analyze small input data that was selected in narrow region on whole ECG, it can continuously scan all ECG data and find important points such as start and end points of P, QRS and T waves within a short time. The star and end points of ECG waves determined by the C-DNNs were compared with the results performed by human experts to estimate the accuracies of the C-DNNs. The C-DNN has 150 inputs, 51 outputs, two hidden layers and one output layer. To find the start and end points, two C-DNNs were trained through deep learning technology and applied to a parameter acquisition algorithms. 12 lead ECG data measured in four patients and obtained through PhysioNet was processed to make training data by human experts. The accuracy of the C-DNNs were evaluated with extra data that were not used at deep learning by comparing the results between C-DNNs and human experts. The averages of the time differences between the C-DNNs and experts were 0.1 msec and 13.5 msec respectively and those standard deviations were 17.6 msec and 15.7 msec. The final step combining the results of C-DNN through the waveforms of 12 leads was successfully determined all 33 waves without error that the time differences of human experts decision were over 20 msec. The reliable decision of the ECG wave's start and end points benefits the acquisition of accurate ECG parameters such as the wave lengths, amplitudes and intervals of P, QRS and T waves.

• Journal of Korea Port Economic Association
• /
• v.33 no.4
• /
• pp.73-82
• /
• 2017

Korea is the fifth largest producer of automobiles in the world, and this industry accounts for the highest portion of the entire manufacturing industry. It is an especially important industry occupying second place in the top 10 export items in Korea. Korea exports about 3 million units of cars produced in the country and abroad, based on new cars and excluding second hand cars. Japan, along with Korea, represents a high portion of the global automobile industry, and it exports more than 4 million cars to the rest of the world. In particular, both Korea and Japan export automobile and used cars produced within the country, almost all of them by PCC(Pure Car Carrier) or PCTC(Pure Car Truck Carrier). Therefore, automobile export ports are located near automobile factories, and are being used in export to foreign countries. However, there are inefficient problems, such as poor port facilities, yard space shortage for loading and unloading operations and lack of proficiency of cargo handling companies. As a result, there are delays in cargo operations, or ships waiting have occurred. Therefore, the purpose of this study is to measure and compare the efficiency of automobile export ports in Korea and Japan. To measure the efficiency of automobile export port, we used CRS and VRS models from DEA. The input and output parameters were set as length of quay, yard area and throughput of cars, and DMUs are 25 ports for evaluating the efficiency. As a result of the efficiency measurements, two Korean ports (Gwangyang and Ulsan) and three Japanes ports (Kanda, Omaezaki, Kanmon-Shimonoseki) showed high efficiency in both models. These results can be used to establish strategies for enhancing efficiency and competitiveness of automobile export ports in Korea and Japan.

• Korean Journal of Ecology and Environment
• /
• v.42 no.2
• /
• pp.242-252
• /
• 2009

The purpose of this study was to develop a Multi-metric Water Quality Assessment (MWQA) model and apply it to dataset sampled from Paldang and Daechung reservoir in 2008. The various water dataset used to this study included 5 year data sets (2003${\sim}$2007) in Korean reservoirs which were obtained from the Ministry of Environment, Korea. In this study, suggested MWQA model has 4 metrics that were composed of 4 parameters such as chemical, physical, biological, and hydrological variables. And, each of the variables attributed total phosphorus (TP) concentration in water, secchi depth (SD) measure in water, chlorophyll-${\alpha}$(Chl-${\alpha}$) concentration in water and the ratio of inflow of water into lakes and efflux of water from lakes, input/output (I/O). First, we established the criteria for trophic boundaries. The boundary between oligotrophic and mesotrophic categories was defined by the lower third of the cumulative distribution of the values. The mesotrophic-eutrophic boundary was defined by the upper third of the distribution. Second, each metric was given by a point-oligo=1, meso=3, eu=5. And then, obtained total score from each metric was divided 5 grade-Excellent, Good, Fair, Poor, and Very poor. As the results of applying the proposed MWQA model, the Paldang reservoir obtained "Fair" or "Poor" grade and Daechung reservoir obtained "Excellent" or "Good" grade. The suggested MWQA model through these procedures will enable to manage efficiently the reservoir. And, more studies such as metric numbers and attributes should be done for the accurate application of the new model.

• Journal of Korea Water Resources Association
• /
• v.41 no.9
• /
• pp.929-941
• /
• 2008

The objectives of this study are to propose a system for combined use of a hydrologic and a hydraulic model for urban flood forecast model and to evaluate the system on the $300km^2$ Jungrang urban watershed area, which is relatively large area as an urban watershed and consequently composed of very complex drainage pipes and streams with different land uses. In this study, SWMM for hydrologic model and HEC-RAS for hydraulic model are used and the study area is divided into 25 subbasins. The SWMM model is used for sewer drainage analysis within each subbasin, while HEC-RAS for unstready flow analysis in the channel streams. Also, this study develops a GUI system composed of mean areal precipitation input component, hydrologic runoff analysis component, stream channel routing component, and graphical representation of model output. The proposed system was calibrated for the model parameters and verified for the model applicability by using the observation data. The correlation coefficients between simulated and observed flows at the 2 important locations were ranged on 0.83-0.98, while the coefficients of model efficiency on 0.60-0.92 for the verification periods. This study also provided the possibilities of manhole overflows and channel bank inundation through the calculated water profile of longitudinal and channel sections, respectively. It can be concluded that the proposed system can be used as a surface runoff and channel routing models for urban flood forecast over the large watershed area.

• Advances in aircraft and spacecraft science
• /
• v.5 no.2
• /
• pp.225-237
• /
• 2018

The difficulties of satellite vibration testing are due to the commonly expressed qualification requirements being incompatible with the limited performance of the entire controlled system (satellite + interface + shaker + controller). Two features cause the problem: firstly, the main satellite modes (i.e., the first structural mode and the high and low tank modes) are very weakly damped; secondly, the controller is just too basic to achieve the expected performance in such cases. The combination of these two issues results in oscillations around the notching levels and high amplitude beating immediately after the mode. The beating overshoots are a major risk source because they can result in the test being aborted if the qualification upper limit is exceeded. Although the abort is, in itself, a safety measure protecting the tested satellite, it increases the risk of structural fatigue, firstly because the abort threshold has been already reached, and secondly, because the test must restart at the same close-resonance frequency and remain there until the qualification level is reached and the sweep frequency can continue. The beat minimum relates only to small successive frequency ranges in which the qualification level is not reached. Although they are less problematic because they do not cause an inadvertent test shutdown, such situations inevitably result in waiver requests from the client. A controlled-system analysis indicates an operating principle that cannot provide sufficient stability: the drive calculation (which controls the process) simply multiplies the frequency reference (usually called cola) and a function of the following setpoint, the ratio between the amplitude already reached and the previous setpoint, and the compression factor. This function value changes at each cola interval, but it never takes into account the sensor signal phase. Because of these limitations, we firstly examined whether it was possible to empirically determine, using a series of tests with a very simple dummy, a controller setting process that significantly improves the results. As the attempt failed, we have performed simulations seeking an optimum adjustment by finding the Least Mean Square of the difference between the reference and response signal. The simulations showed a significant improvement during the notch beat and a small reduction in the beat amplitude. However, the small improvement in this process was not useful because it highlighted the need to change the reference at each cola interval, sometimes with instructions almost twice the qualification level. Another uncertainty regarding the consequences of such an approach involves the impact of differences between the estimated model (used in the simulation) and the actual system. As limitations in the current controller were identified in different approaches, we considered the feasibility of a new controller that takes into account an estimated single-input multi-output (SIMO) model. Its parameters were estimated from a very low-level throughput. Against this backdrop, we analyzed the feasibility of an LQG control in cancelling beating, and this article highlights the relevance of such an approach.