• 대한전자공학회논문지SP
• /
• 제39권5호
• /
• pp.486-493
• /
• 2002

본 논문에서는 웨이블릿 영역에서 각 부밴드에 대한 영역별 대역간 양방향 예측과 확장된 SPIHT (set partition in hierarchical trees)를 이용한 효율적인 인공위성 다분광 화상데이터의 압축 방법을 제안하였다. 이 방법에서는 가시광선 영역과 적외선 영역에서 다른 대역과 분광적 상관성이 큰 대역을 기준대역 (feature band)으로 각각 결정하고, 이 대역들에 대해 웨이블릿 변환 (wavelet transform, WT)을 행한 후 SPIHT를 행하여 부호화함으로써 대역내 (intraband) 중복성을 제거한다. 기준대역과 대역간 상관성이 큰 예측대역 (prediction band)들에 대해서는 웨이블릿 변환을 행한 후, 각 대역의 기저밴드의 대역별 특성을 이용하여 영역분류를 하고, 각 부밴드에 대한 영역별 대역간 양방향 예측 (classified interband bidirec- tional prediction)을 행함으로써 대역간 (interband) 중복성을 제거하여 압축 효율을 향상시킨다. 또한 확장된 SPIHT의 부호화 효율을 높이기 위해 예측오차의 최대값에 따라 재배열된 대역들에 대해 확장된 SPIHT를 행하여 예측오차를 부호화함으로써, 예측에 따른 오차를 보상하여 화질을 향상시킨다. 실제 다분광 화상데이터에 대한 모의 실험을 통하여 제안한 방법의 부호화 효율이 기존의 방법에 비하여 우수함을 확인하였다.

• Korean Chemical Engineering Research
• /
• 제61권3호
• /
• pp.388-393
• /
• 2023

지도 학습 기반의 신경 망을 활용한 공학적 자료의 분석은 화학공학 공정 최적화, 미세 먼지 농도 추정, 열역학적 상평형 예측, 이동 현상 계의 물성 예측 등 다양한 분야에서 활용되고 있다. 신경 망의 지도 학습은 학습 자료를 요구하며, 주어진 학습 자료의 구성에 따라 학습 성능이 영향을 받는다. 빈번히 관찰되는 공학적 자료 중에는 DNA의 길이, 분석 물질의 농도 등과 같이 로그 간격으로 주어지는 자료들이 존재한다. 본 연구에서는 넓은 범위에 분포된 로그 간격의 학습 자료를 기계 학습으로 처리하는 경우, 사용 가능한 손실 함수들의 학습 성능을 정량적으로 평가하였으며, 적합한 학습 자료 구성 방식을 연구하였다. 이를 수행하고자, 100×100의 가상 이미지를 활용하여 기계 학습의 회귀 과업을 구성하였다. 4개의 손실 함수들에 대하여 (i) 오차 행렬, (ii) 최대 상대 오차, (iii) 평균 상대 오차로 정량적 평가하여, mape 혹은 msle가 본 연구에서 다룬 과업에 대해 최적의 손실 함수가 됨을 알아내었다. 또한, 학습 자료의 값이 넓은 범위에 걸쳐 분포하는 경우, 학습 자료의 구성을 로그 간격 등을 고려하여 균등 선별하는 방식이 높은 학습 성능을 보임을 밝혀내었다. 본 연구에서 다룬 회귀 과업은 DNA의 길이 예측, 생체 유래 분자 분석, 콜로이드 용액의 농도 추정 등의 공학적 과업에 적용 가능하며, 본 결과를 활용하여 기계 학습의 성능과 학습 효율의 증대를 기대할 수 있을 것이다.

• Journal of Biosystems Engineering
• /
• 제37권3호
• /
• pp.166-176
• /
• 2012

Purpose: This study was conducted to investigate the potential of interactance mode of NIR spectroscopy technology for the estimation of soluble solids content (SSC) and firmness of muskmelons. Methods: Melon samples were taken from local greenhouses in three different harvesting seasons (experiments 1, 2, and 3). The fruit attributes were measured at the 6 points on an equator of each sample where the spectral data were collected. The prediction models were developed using the original spectral data and the spectral data sets preprocessed by 20 methods. The performance of the models was compared. Results: In the prediction of SSC, the highest coefficient of determination ($R_{cv}{^2}$) values of the cross-validation was 0.755 (standard error of prediction, SEP=$0.89^{\circ}Brix$) with the preprocessing of normalization with range in experiment 1. The highest coefficient of determination in the robustness tests, $R_{rt}{^2}$=0.650 (SEP=$1.03^{\circ}Brix$), was found when the best model of experiment 3 was evaluated with the data set of experiment 2. The best $R_{cv}{^2}$ for the prediction of firmness was 0.715 (SEP=3.63 N) when no preprocessing was applied in experiment 1. The highest $R_{rt}{^2}$ was 0.404 (SEP=5.30 N) when the best model of experiment 3 was applied to the data set of experiment 1. Conclusions: From the test results, it can be concluded that the interactance mode of VIS/NIR spectroscopy technology has a great potential to measure SSC and firmness of thick-skinned muskmelons.

• 문화기술의 융합
• /
• 제9권3호
• /
• pp.763-768
• /
• 2023

본 논문에서는 의약품 유통량 예측을 위해 기존의 통계 방식(ARIMA)과 머신러닝 방식(Informer)을 개발하고 비교하였다. 일별 데이터의 예측에서는 머신러닝 기반의 모델이 유리하며, 월별 예측에서는 ARIMA를 활용하고 데이터가 증가하면서 Informer로 전환하는 것이 효과적임을 발견하였다. 예측 에러율(RMSE)은 기존 방식 대비 26.6% 낮아졌으며, 예측 정확도도 13% 개선되어 86.2%의 결과를 보였다. 본 논문을 통해 통계적 방법과 머신러닝 방법을 앙상블하여 최상의 결과를 얻을 수 있다는 장점을 발견하였다. 또한 머신러닝 기반의 AI 모델은 불규칙한 상황에서도 딥러닝 연산을 통해 최선의 결과를 도출할 수 있으며, 상용화 이후에는 데이터양이 증가함에 따라 성능이 향상될 것으로 기대된다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제11권6호
• /
• pp.3029-3045
• /
• 2017

The main processes of a cognitive radio system include spectrum sensing, spectrum decision, spectrum sharing, and spectrum conversion. Experimental results show that these stages introduce a time delay that affects the spectrum sensing accuracy, reducing its efficiency. To reduce the time delay, the frequency spectrum prediction was proposed to alleviate the burden on the spectrum sensing. In this paper, the deep recurrent neural network (DRNN) was proposed to predict the spectrum of multiple time slots, since the existing methods only predict the spectrum of one time slot. The continuous state of a channel is divided into a many time slots, forming a time series of the channel state. Since there are more hidden layers in the DRNN than in the RNN, the DRNN has fading memory in its bottom layer as well as in the past input. In addition, the extended Kalman filter was used to train the DRNN, which overcomes the problem of slow convergence and the vanishing gradient of the gradient descent method. The spectrum prediction based on the DRNN was verified with a WiFi signal, and the error of the prediction was analyzed. The simulation results proved that the multiple slot spectrum prediction improved the spectrum efficiency and reduced the energy consumption of spectrum sensing.

• 한국도로학회논문집
• /
• 제14권5호
• /
• pp.133-143
• /
• 2012

PURPOSES: Using the collected data for crash, traffic volume, and design elements on ramps between 2007 and 2009, this research effort was initiated to develop traffic crash prediction models for expressway ramps. METHODS: Three negative binomial regression models and three zero-inflated negative binomial regression models were developed for individual ramp types, including direct, semi-direct and loop, respectively. For validating the developed models, authors compared the estimated crash frequencies with actual crash frequencies of twelve randomly selected interchanges, the ramps of which have not been used for model developing. RESULTS: The results show that the negative binomial regression models for direct, semi-direct and loop ramps showed 60.3%, 63.8% and 48.7% error rates on average whereas the zero-inflated negative binomial regression models showed 82.1%, 120.4% and 57.3%, respectively. CONCLUSIONS: Conclusively, the negative binomial regression models worked better in traffic crash prediction than the zero-inflated negative binomial regression models for estimating the frequency of traffic accidents on expressway ramps.

• Structural Engineering and Mechanics
• /
• 제86권2호
• /
• pp.181-195
• /
• 2023

To predict the rheological behaviours along with the compressive strength of self-compacting concrete that incorporates environmentally friendly ingredients as cement substitutes, a comparative evaluation of machine learning methods is conducted. To model four parameters, slump flow diameter, L-box ratio, V-funnel time, as well as compressive strength at 28 days-a complete mix design dataset from available pieces of literature is gathered and used to construct the suggested machine learning standards, SVM, MARS, and Mp5-MT. Six input variables-the amount of binder, the percentage of SCMs, the proportion of water to the binder, the amount of fine and coarse aggregates, and the amount of superplasticizer are grouped in a particular pattern. For optimizing the hyper-parameters of the MARS model with the lowest possible prediction error, a gravitational search algorithm (GSA) is required. In terms of the correlation coefficient for modelling slump flow diameter, L-box ratio, V-funnel duration, and compressive strength, the prediction results showed that MARS combined with GSA could improve the accuracy of the solo MARS model with 1.35%, 11.1%, 2.3%, as well as 1.07%. By contrast, Mp5-MT often demonstrates greater identification capability and more accurate prediction in comparison to MARS-GSA, and it may be regarded as an efficient approach to forecasting the rheological behaviors and compressive strength of SCC in infrastructure practice.

• 한국소음진동공학회논문집
• /
• 제22권1호
• /
• pp.9-14
• /
• 2012

This paper investigates the prediction of the dynamic strain response using acceleration response only. Two methods are proposed for the strain prediction; one is based on beam theory and the other is calculated by the frequency response function between acceleration and strain. First, it is estimated the dynamics of the simple notched beam, including the non-linearity, through the uni-axial vibration testing. Then, the dynamic strain response is predicted under two different methods using acceleration response. The validation of proposed methods is conducted by the comparison between measured strain and predicted values. The comparison reveals that the proposed method based on the FRF between acceleration and strain is more reliable one than that stemmed from beam theory and the maximum relative error is less than 8 %.

• Computers and Concrete
• /
• 제27권4호
• /
• pp.305-317
• /
• 2021

The bond between the concrete and bar is a main factor affecting the performance of the reinforced concrete (RC) members, and since the steel corrosion reduces the bond strength, studying the bond behavior of concrete and GFRP bars is quite necessary. In this research, a database including 112 concrete beam test specimens reinforced with spliced GFRP bars in the splitting failure mode has been collected and used to estimate the concrete-GFRP bar bond strength. This paper aims to accurately estimate the bond strength of spliced GFRP bars in concrete beams by applying three soft computing models including multivariate adaptive regression spline (MARS), Kriging, and M5 model tree. Since the selection of regularization parameters greatly affects the fitting of MARS, Kriging, and M5 models, the regularization parameters have been so optimized as to maximize the training data convergence coefficient. Three hybrid model coupling soft computing methods and genetic algorithm is proposed to automatically perform the trial and error process for finding appropriate modeling regularization parameters. Results have shown that proposed models have significantly increased the prediction accuracy compared to previous models. The proposed MARS, Kriging, and M5 models have improved the convergence coefficient by about 65, 63 and 49%, respectively, compared to the best previous model.

• Geomechanics and Engineering
• /
• 제30권1호
• /
• pp.75-91
• /
• 2022

This paper aims to show how to use several Machine Learning (ML) methods to estimate the TBM penetration rate systematically (TBM-PR). To this end, 1125 datasets including uniaxial compressive strength (UCS), Brazilian tensile strength (BTS), punch slope index (PSI), distance between the planes of weakness (DPW), orientation of discontinuities (alpha angle-α), rock fracture class (RFC), and actual/measured TBM-PRs were established. To evaluate the ML methods' ability to perform, the 5-fold cross-validation was taken into consideration. Eventually, comparing the ML outcomes and the TBM monitoring data indicated that the ML methods have a very good potential ability in the prediction of TBM-PR. However, the long short-term memory model with a correlation coefficient of 0.9932 and a route mean square error of 2.68E-6 outperformed the remaining six ML algorithms. The backward selection method showed that PSI and RFC were more and less significant parameters on the TBM-PR compared to the others.