• 상하수도학회지
• /
• 제12권4호
• /
• pp.43-52
• /
• 1998

The purpose of this study is to establish a method of estimating the daily urban water demend using Backpropagation algorithm is part of ANN(Artificial Neural Network). This method will be used for the development of the efficient management and operations of the water supply facilities. The data used were the daily urban water demend, the population and weather conditions such as treperarture, precipitation, relative humidity, etc. Kwangju city was selected for the case study area. We adjusted the weights of ANN that are iterated the training data patterns. We normalized the non-stationary time series data [-1,+1] to fast converge, and choose the input patterns by statistical methods. We separated the training and checking patterns form input date patterns. The performance of ANN is compared with multiple-regression method. We discussed the representation ability the model building process and the applicability of ANN approach for the daily water demand. ANN provided the reasonable results for time series forecasting.

• Structural Engineering and Mechanics
• /
• 제34권2호
• /
• pp.159-174
• /
• 2010

Elastic buckling load of perforated steel plates is typically predicted using the finite element or conjugate load/displacement methods. In this paper an artificial neural network (ANN)-based formula is presented for the prediction of the elastic buckling load of rectangular plates having a circular cutout. By using this formula, the elastic buckling load of perforated plates can be calculated easily without setting up an ANN platform. In this study, the center of a circular cutout was chosen at different locations along the longitudinal x-axis of plates subjected to linearly varying loading. The results of the finite element method (FEM) produced by the commercial software package ANSYS are used to train and test the network. The accuracy of the proposed formula based on the trained ANN model is evaluated by comparing with the results of different researchers. The results show that the presented ANN-based formula is practical in predicting the elastic buckling load of perforated plates without the need of an ANN platform.

• 한국수자원학회논문집
• /
• 제54권spc1호
• /
• pp.1143-1154
• /
• 2021

활성화함수의 선택은 인공신경망(Artificial Neural Network, ANN) 모델의 지하수위 예측성능에 큰 영향을 미친다. 특히 제주도의 중산간 지역과 같이 지하수위의 변동폭이 크고 변동양상이 복잡한 경우 적절한 지하수위 예측을 위해서는 다양한 활성화함수의 비교분석을 통한 최적의 활성화함수 선택이 반드시 필요하다. 본 연구에서는 지하수위의 변동폭이 크고 변동양상이 복잡한 제주도 표선유역 중산간지역 2개 지하수위 관측정을 대상으로 5개의 활성화함수(sigmoid, hyperbolic tangent (tanh), Rectified Linear Unit (ReLU), Leaky Rectified Linear Unit (Leaky ReLU), Exponential Linear Unit (ELU))를 ANN 모델에 적용하여 지하수위 예측결과를 비교 및 분석하고 최적 활성화함수를 도출하였다. 그리고 최근 널리 사용되고 있는 순환신경망 모델인 Long Short-Term Memory (LSTM) 모델의 결과와 비교분석하였다. 분석결과 지하수위 변동폭이 상대적으로 큰 관측정과 상대적으로 작은 관측정에 대한 지하수위 예측에 대해서는 각각 ELU와 Leaky ReLU 함수가 최적의 활성화함수로 도출되었다. 반면 sigmoid 함수는 학습기간에 대해 5개 활성화함수 중 예측성능이 가장 낮았으며 첨두 및 최저 지하수위 예측에서 적절하지 못한 결과를 도출하였다. 따라서 ANN-sigmoid 모델은 가뭄기간의 지하수위 예측을 통한 지하수자원 관리목적으로 사용할 경우 주의가 필요하다. ANN-ELU와 ANN-Leaky ReLU 모델은 LSTM 모델과 대등한 지하수위 예측성능을 보여 활용가능성이 충분히 있으며 LSTM 모델은 ANN 모델들 보다 예측성능이 높아 인공지능 모델의 예측성능 비교분석 시 참고 모델로 활용될 수 있다. 마지막으로 학습기간의 정보량에 따라 학습기간의 지하수위 예측성능이 검증 및 테스트 기간의 예측성능보다 낮을 수 있다는 것을 확인하였으며, 관측지하수위의 변동폭이 크고 변동양상이 복잡할수록 인공지능 모델별 지하수위 예측능력의 차이는 커졌다. 본 연구에서 제시한 5개의 활성화함수를 적용한 연구방법 및 비교분석 결과는 지하수위 예측뿐만 아니라 일단위 하천유출량 및 시간단위 홍수량 등 지표수 예측을 포함한 다양한 연구에 유용하게 사용될 수 있다.

• 한국관개배수논문집
• /
• 제11권2호
• /
• pp.95-103
• /
• 2004

The objective of this study is to simulate streamflow and water quality for Gyeongan watershed (561.1 $km^2$) using AnnAGNPS (Bingner et al., 2000). The model was calibrated and verified for three years (2000, 2002, 2003) stream discharge and w

• 한국항해항만학회지
• /
• 제32권8호
• /
• pp.589-596
• /
• 2008

선박의 접안운동을 자동화하기 위하여 인공신경망(Artificial Neural Network, 이하 ANN)에 의한 제어를 수행하였다. ANN은 시스템의 비선형성이 표현 가능하므로 접안운동과 같은 비선형성이 강한 조종운동에 적합하다. 입력층과 출력층 사이에 하나 이상의 중간층이 존재하는 다층 인식자(Multi-layer perceptron)를 사용하였고, 교사 데이터(Teaching data)와 역전파(Back-Propagation) 알고리즘을 사용하여 신경망의 출력값과 목표 출력값 사이의 오차가 최소가 되도록 신경망 학습을 수행하였다. 접안 시 저속조종 수학모델을 사용하여 접안 시뮬레이션을 수행하였으며, ANN의 입력층 성분(unit)이 8개인 구조와 6개인 구조의 접안 제어를 비교하였다. 시뮬레이션 결과, 두 ANN에 의하여 접안 경로 선택에 차이가 나타났으나 접안 조건은 모두 만족하였다.

• 소성∙가공
• /
• 제29권2호
• /
• pp.76-88
• /
• 2020

The need for advanced high strength steel (AHSS) forming technology is increasing as interest in light weight and safe automobiles increases. Multipoint dieless forming (MDF) is a novel sheet metal forming technology that can create any desired longitudinal and transverse curvature in sheet metal. However, since the springback phenomenon becomes larger with high strength metal such as AHSS, predicting the required MDF to produce the exact desired curvature in two directions is more difficult. In this study, a prediction model using artificial neural network (ANN) was developed to predict the springback that occurs during AHSS forming through MDF. In order to verify the validity of model, a fit test was performed and the results were compared with the conventional regression model. The data required for training was obtained through simulation, then further random sample data was created to verify the prediction performance. The predicted results were compared with the simulation results. As a result of this comparison, it was found that the prediction of our ANN based model was more accurate than regression analysis. If a sufficient amount of data is used in training, the ANN model can play a major role in reducing the forming cost of high-strength steels.

• Steel and Composite Structures
• /
• 제49권6호
• /
• pp.633-643
• /
• 2023

In this study, the natural frequencies of Functional Graded Materials (FGM) plates are predicted using Artificial Neural Network (ANN). A model based on Third-order Shear Deformation Theory (TSDT) and FEM is used to train the ANN model. Different training methods are tested to simulate input and output dependency. As this is a parametric model, several architectures and optimization algorithms were tested. The proposed model allows us to minimize the CPU time to evaluate candidate material properties for FGM plate material selection and demonstrate their influence on dynamic behavior. Consequently, the time required for the FGM design process (candidate materials for material selection) and the geometric optimization of the FGM structure would remain reasonable. The ANN model can help industries to produce FGM plates with good mechanical properties of the selected materials. I addition, this model can be used to directly predict vibration behavior by testing a large number of FGM plates, representing all possible combinations of metals and ceramics in today's industry, without having to solve any eigenvalue problems.

• 방사성폐기물학회지
• /
• 제2권1호
• /
• pp.35-40
• /
• 2004

Final disposal of radioactive waste generated from Nuclear Power Plant (NPP) requires the detailed information about the characteristics and the quantities of radionuclides in waste package. Most of these radionuclides are difficult to measure and expensive to assay. Thus it is suggested to the indirect method by which the concentration of the Difficult-to-Measure (DTM) nuclide is estimated using the correlations of concentration - it is called the scaling factor - between Easy-to-Measure (Key) nuclides and DTM nuclides with the measured concentration of the Key nuclide. In general, the scaling factor is determined by the log mean average (LMA) method and the regression method. However, these methods are inadequate to apply to fission product nuclides and some activation product nuclides such as 14$^{C}$ and 90$^{Sr}$ . In this study, the artificial neural network (ANN) method is suggested to improve the conventional SF determination methods - the LMA method and the regression method. The root mean squared errors (RMSE) of the ANN models are compared with those of the conventional SF determination models for 14$^{C}$ and 90$^{Sr}$ in two parts divided by a training part and a validation part. The SF determination models are arranged in the order of RMSEs as the following order: ANN model

• Geomechanics and Engineering
• /
• 제22권5호
• /
• pp.397-414
• /
• 2020

In this paper, practical predictive models for soil shear strength parameters are proposed. As cohesion and internal friction angle are of essential shear strength parameters in any geotechnical studies, we try to predict them via artificial neural network (ANN) and neuro-imperialism approaches. The proposed models was based on the result of a series of consolidated undrained triaxial tests were conducted on reinforced sandy soil. The experimental program surveys the increase in internal friction angle of sandy soil due to addition of polypropylene fibers with different lengths and percentages. According to the result of the experimental study, the most important parameters impact on internal friction angle i.e., fiber percentage, fiber length, deviator stress, and pore water pressure were selected as predictive model inputs. The inputs were used to construct several ANN and neuro-imperialism models and a series of statistical indices were calculated to evaluate the prediction accuracy of the developed models. Both simulation results and the values of computed indices confirm that the newly-proposed neuro-imperialism model performs noticeably better comparing to the proposed ANN model. While neuro-imperialism model has training and test error values of 0.068 and 0.094, respectively, ANN model give error values of 0.083 for training sets and 0.26 for testing sets. Therefore, the neuro-imperialism can provide a new applicable model to effectively predict the internal friction angle of fiber-reinforced sandy soil.

• 한국수자원학회논문집
• /
• 제45권12호
• /
• pp.1321-1330
• /
• 2012

본 연구에서는 비선형적 모델인 웨이블렛-인공신경망을 적용하여 충주댐 유역의 일유입량을 예측하였다. 일반적으로 시계열 자료는 경향성, 주기성 및 추계학적 성분의 선형조합으로 이루어져 있다. 그러나 이러한 자료를 통해 시계열 모형 구축 시 경향성 및 주기성은 제거되어야하는 성분이다. 따라서 수문기상자료에 포함되어있는 경향성 및 주기성과 같은 비선형 동역학적 잡음과 측정과정에서 발생하는 단순잡음을 제거시키기 위해 디노이징기법인 웨이블렛 변환을 적용하였다. 웨이블렛 변환을 적용한 자료를 입력자료로 사용한 웨이블렛-인공신경망(WANN)과 원자료를 사용한 인공신경망(ANN)을비교하였다. 산정결과 결정계수와 선형회귀를 통한 기울기는 WANN이 ANN보다 각각0.032, 0.0115 더 큰값을 나타냈고, 타겟값과 예측값 사이의 오차를 나타내는 RMSE와 RRMSE는 WANN 모형이 ANN 보다 각각 37.388, 0.099 더 작은값을 나타냈다. 따라서 본 연구에서 적용한 WANN 모형이 ANN 보다 정확한 결과를 나타내었으며, 웨이블렛 변환을 통한 디노이징 기법의 적용이 잡음이 포함되어 있는 원자료의 사용보다 더 정확한 예측을 하는 것으로 판단된다.