• Proceedings of the Korea Water Resources Association Conference
• /
• 2017.05a
• /
• pp.392-395
• /
• 2017

본 연구에서는 일 증발접시 증발량 산정을 위한 딥러닝 (deep learning) 모형의 적용성을 평가하였다. 본 연구에서 적용된 딥러닝 모형은 deep belief network (DBN) 기반 deep neural network (DNN) (DBN-DNN) 모형이다. 모형 적용성 평가를 위하여 부산 관측소에서 측정된 기상자료를 활용하였으며, 증발량과의 상관성이 높은 기상변수들 (일사량, 일조시간, 평균지상온도, 최대기온)의 조합을 고려하여 입력변수집합 (Set 1, Set 2, Set 3)별 모형을 구축하였다. DBN-DNN 모형의 성능은 통계학적 모형성능 평가지표 (coefficient of efficiency, CE; coefficient of determination, $r^2$; root mean square error, RMSE; mean absolute error, MAE)를 이용하여 평가되었으며, 기존의 두가지 형태의 ANN (artificial neural network), 즉 모형학습 시 SGD (stochastic gradient descent) 및 GD (gradient descent)를 각각 적용한 ANN-SGD 및 ANN-GD 모형과 비교하였다. 효과적인 모형학습을 위하여 각 모형의 초매개변수들은 GA (genetic algorithm)를 이용하여 최적화하였다. 그 결과, Set 1에 대하여 ANN-GD1 모형, Set 2에 대하여 DBN-DNN2 모형, Set 3에 대하여 DBN-DNN3 모형이 가장 우수한 모형 성능을 나타내는 것으로 분석되었다. 비록 비교 모형들 사이의 모형성능이 큰 차이를 보이지는 않았으나, 모든 입력집합에 대하여 DBN-DNN3, DBN-DNN2, ANN-SGD3 순으로 모형 효율성이 우수한 것으로 나타났다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2019.05a
• /
• pp.123-123
• /
• 2019

최근 기후변화로 인해 중국, 한국, 일본, 몽골 등을 포함한 동아시아 지역은 태풍, 가뭄, 홍수와 같은 자연재해의 발생 빈도가 증가하고 있는 추세이다. 중국의 경우 2017년 극심한 가뭄으로 1,850만 (ha)의 농작물 피해가 발생하였으며, 몽골 또한 2017년 4월 이후 극심한 가뭄으로 사막화가 급속도로 진행되고 있다. 위성 기반의 강우 자료는 공간과 시간 해상도가 높아짐에 따라 지상관측소 강수량 자료의 대체 수단으로 이용되고 있다. 본 연구에서는 Climate Hazards Groups InfraRed Precipitation with Station (CHIRPS), Precipitation Estimation From Remotely Sensed Information Using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR), Global Precipitation Climatology Centre (GPCC) 강우 위성 자료를 활용하여 기상학적 가뭄지수인 표준강수지수 (Standardized Precipitation Index, SPI)를 산정하였다. 시간 해상도는 월별 영상을 기준으로 2008년부터 2017년까지 지난 10년간의 데이터를 이용하였으며, 각각 격자가 다른 위성영상을 기존 기상관측소와 비교하였다. 피어슨 상관계수 (Pearson Correlation Coefficient, R)를 활용하여 강우 위성 영상과 지상관측소의 상관관계를 분석하고, 평균절대오차 (Mean Absolute Error, MAE), 평균제곱근오차 (Root Mean Square Error, RMSE)를 통해 통계적으로 정확도를 분석하였다. 인공위성 강수량 자료는 미계측 지역이 많은 곳이나 측정이 불가능한 지역에 효율성 측면에서 중요한 이점을 제공할 것으로 판단된다.

• Steel and Composite Structures
• /
• v.39 no.4
• /
• pp.471-491
• /
• 2021

In this study, we estimate the ultimate load of rectangular concrete-filled steel tubes (CFST) by developing a novel hybrid predictive model (ANN-BCMO) which is a combination of balancing composite motion optimization (BCMO) - a very new optimization technique and artificial neural network (ANN). For this aim, an experimental database consisting of 422 datasets is used for the development and validation of the ANN-BCMO model. Variables in the database are related with the geometrical characteristics of the structural members, and the mechanical properties of the constituent materials (steel and concrete). Validation of the hybrid ANN-BCMO model is carried out by applying standard statistical criteria such as root mean square error (RMSE), coefficient of determination (R²), and mean absolute error (MAE). In addition, the selection of appropriate values for parameters of the hybrid ANN-BCMO is conducted and its robustness is evaluated and compared with the conventional ANN techniques. The results reveal that the new hybrid ANN-BCMO model is a promising tool for prediction of the ultimate load of rectangular CFST, and prove the effective role of BCMO as a powerful algorithm in optimizing and improving the capability of the ANN predictor.

• Computers and Concrete
• /
• v.27 no.3
• /
• pp.225-239
• /
• 2021

Carbon fiber reinforced polymer (CFRP) has extensive use in strengthening reinforced concrete structures due to its high strength and elastic modulus, low weight, fast and easy application, and excellent durability performance. Many studies have been carried out to determine the performance of the CFRP confined concrete cylinder. Although studies about the prediction of confined compressive strength using ANN are in the literature, the insufficiency of the studies to predict the strain of confined concrete cylinder using ANN, which is the most appropriate analysis method for nonlinear and complex problems, draws attention. Therefore, to predict both strengths and also strain values, two different ANNs were created using an extensive experimental database. The strength and strain networks were evaluated with the statistical parameters of correlation coefficients (R2), root mean square error (RMSE), and mean absolute error (MAE). The estimated values were found to be close to the experimental results. Mathematical equations to predict the strength and strain values were derived using networks prepared for convenience in engineering applications. The sensitivity analysis of mathematical models was performed by considering the inputs with the highest importance factors. Considering the limit values obtained from the sensitivity analysis of the parameters, the performances of the proposed models were evaluated by using the test data determined from the experimental database. Model performances were evaluated comparatively with other analytical models most commonly used in the literature, and it was found that the closest results to experimental data were obtained from the proposed strength and strain models.

• Smart Media Journal
• /
• v.11 no.5
• /
• pp.38-47
• /
• 2022

Honey is one of the most significant ingredients in conventional food production in different regions of the world. Honey is commonly used as an ingredient in ethnic food. Beekeeping is performed in various locations as part of the local food culture and an occupation related to pollinator production. It is important to conduct beekeeping so that it generates food culture and helps regulate the regional environment in an integrated manner in preserving and improving local food culture. This study analyzes different types of environmental factors of a smart bee farm. The major goal of this study is to determine the best prediction model between the linear regression model (LM) and the support vector regression model (SVR) based on the environmental factors of a smart bee farm. The performance of prediction models is measured by R² value, root mean squared error (RMSE), and mean absolute error (MAE). From all analysis reports, the best prediction model is the support vector regression model (SVR) with a low coefficient of variation, and the R² values for Farm inside temperature, bee box inside temperature, and Farm inside humidity are 0.97, 0.96, and 0.44.

• Journal of information and communication convergence engineering
• /
• v.20 no.4
• /
• pp.303-308
• /
• 2022

Simulating the heat transfer in a composite material is an important topic in material science. Difficulties arise from the fact that adjacent materials cannot match perfectly, resulting in discontinuity in the temperature variables. Although there have been several numerical methods for solving the heat-transfer problem in imperfect contact conditions, the methods known so far are complicated to implement, and the computational times are non-negligible. In this study, we developed a ResNet-type deep neural network for simulating a heat transfer model in a composite material. To train the neural network, we generated datasets by numerically solving the heat-transfer equations with Kapitza thermal resistance conditions. Because datasets involve various configurations of composite materials, our neural networks are robust to the shapes of material-material interfaces. Our algorithm can predict the thermal behavior in real time once the networks are trained. The performance of the proposed neural networks is documented, where the root mean square error (RMSE) and mean absolute error (MAE) are below 2.47E-6, and 7.00E-4, respectively.

• Computers and Concrete
• /
• v.34 no.4
• /
• pp.477-487
• /
• 2024

The in-situ stresses of concrete are an essential index for assessing the safety performance of concrete structures. Conventional methods for pore pressure release often face challenges in selecting drilling ring parameters, uncontrollable stress release, and unstable detection accuracy. In this paper, the parameters affecting the results of the concrete ring hole stress release method are cross-combined, and finite elements are used to simulate the combined parameters and extract the stress release values to establish a training set. The GridSearchCV function is utilized to determine the optimal hyperparameters. The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are used as evaluation indexes to train the gradient boosting decision tree (GBDT) algorithm, and the other three common algorithms are compared. The RMSE of the GBDT algorithm for the test set is 4.499, and the R² of the GBDT algorithm for the test set is 0.962, which is 9.66% higher than the R² of the best-performing comparison algorithm. The model generated by the GBDT algorithm can accurately calculate the concrete in-situ stresses based on the drilling ring parameters and the corresponding stress release values and has a high accuracy and generalization ability.

• Structural Engineering and Mechanics
• /
• v.81 no.5
• /
• pp.565-574
• /
• 2022

The infrastructures such as offshore, bridges, power plant, oil and gas piping and aircraft operate in a harsh environment during their service life. Structural integrity of engineering components used in these industries is paramount for the reliability and economics of operation. Two regression models based on the concept of Gaussian process regression (GPR) and Minimax probability machine regression (MPMR) were developed to predict stress intensity factor range (𝚫K). Both GPR and MPMR are in the frame work of probability distribution. Models were developed by using the fatigue crack growth data in MATLAB by appropriately modifying the tools. Fatigue crack growth experiments were carried out on Eccentrically-loaded Single Edge notch Tension (ESE(T)) specimens made of API 5L X65 Grade steel in inert and corrosive environments (2.0% and 3.5% NaCl). The experiments were carried out under constant amplitude cyclic loading with a stress ratio of 0.1 and 5.0 Hz frequency (inert environment), 0.5 Hz frequency (corrosive environment). Crack growth rate (da/dN) and stress intensity factor range (𝚫K) values were evaluated at incremental values of loading cycle and crack length. About 70 to 75% of the data has been used for training and the remaining for validation of the models. It is observed that the predicted SIF range is in good agreement with the corresponding experimental observations. Further, the performance of the models was assessed with several statistical parameters, namely, Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Efficiency (E), Root Mean Square Error to Observation's Standard Deviation Ratio (RSR), Normalized Mean Bias Error (NMBE), Performance Index (ρ) and Variance Account Factor (VAF).

• Journal of Korea Port Economic Association
• /
• v.39 no.2
• /
• pp.129-141
• /
• 2023

This study analyzes the dynamic characteristics of daily freight rates of dry bulk and tanker shipping markets and their forecasting accuracy by using the error correction models. In order to calculate the error terms from the co-integrated time series, this study uses the common stochastic trend model (CSTM model) and vector error correction model (VECM model). First, the error correction model using the error term from the CSTM model yields more appropriate results of adjustment speed coefficient than one using the error term from the VECM model. Furthermore, according to the adjusted determination coefficients (adjR²), the error correction model of CSTM-model error term shows more model fitness than that of VECM-model error term. Second, according to the criteria of mean absolute error (MAE) and mean absolute scaled error (MASE) which measure the forecasting accuracy, the results show that the error correction model with CSTM-model error term produces more accurate forecasts than that of VECM-model error term in the 12 cases among the total 15 cases. This study proposes the analysis and forecast tasks 1) using both of the CSTM-model and VECM-model error terms at the same time and 2) incorporating additional data of commodity and energy markets, and 3) differentiating the adjustment speed coefficients based the sign of the error term as the future research topics.

• Journal of Korea Water Resources Association
• /
• v.56 no.2
• /
• pp.91-101
• /
• 2023

In this study, we estimated the concentration of chlorophyll-a (Chl-a) using hyperspectral water surface reflectance in an inland weir (Baekjae weir) and estuarine reservoir (Namyang Reservoir) for monitoring the occurrence of algae in freshwater in South Korea. The hyperspectral reflectance was measured by aircraft in Baekjae Weir (BJW) from 2016 to 2017, and a drone in Namyang Reservoir (NYR) from 2020 to 2021. The 30 reflectance bands (BJW: 400-530, 620-680, 710-730, 760-790 nm, NYR: 400-430, 655-680, 740-800 nm) that were highly related to Chl-a concentration were selected using permutation importance. Artificial neural network based Chl-a estimation model was developed using the selected reflectance in both water bodies. And the performance of the model was evaluated with the coefficient of determination (R²), the root mean square error (RMSE), and the mean absolute error (MAE). The performance evaluation results of the Chl-a estimation model for each watershed was R²: 0.63, 0.82, RMSE: 9.67, 6.99, and MAE: 11.25, 8.48, respectively. The developed Chl-a model of this study may be used as foundation tool for the optimal management of freshwater algal blooms in the future.