• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 1997.10a
• /
• pp.177-180
• /
• 1997

Outliers, unexpected extraordinary observations that look discordant from most observation in a data set are commonplace in various kinds of data analysis. Since the effect of outliers on model identification could be serious, the aim of this paper is to present some ways of handling outliers in given data set and to specify a model in the presence of outliers. A procedure based on neural network which identifies outliers, removes their effects, and specifies a model for the underlying process is proposed. In contrast with traditional parametric methods requiring to estimate the model's structure and parameters before detecting outliers, the proposed procedure is a nonparametric method without the estimation of model's structure and parameters before handling outliers and could be applied for real problems in the presence of outliers. The proposed methodology is performed as followings. Firstly, outliers are detected and the detected outliers replace the prediction values using outliers detection neural network. The data set removing the effect of outliers is retraining using neural network. Therefore the effects of outliers are removed and the modeling precision can be improved. Experimental results show that the proposed method is suitable for predicting data set in the presence of outliers.

• Computers and Concrete
• /
• v.21 no.1
• /
• pp.47-54
• /
• 2018

Artificial neural network models can be successfully used to simulate the complex behavior of many problems in civil engineering. As compared to conventional computational methods, this popular modeling technique is powerful when the relationship between system parameters is intrinsically nonlinear, or cannot be explicitly identified, as in the case of concrete behavior. In this investigation, an artificial neural network model was developed to assess the residual compressive strength of self-compacted concrete at elevated temperatures ($20-900^{\circ}C$) and various relative humidity conditions (28-99%). A total of 332 experimental datasets, collected from available literature, were used for model calibration and verification. Data used in model development incorporated concrete ingredients, filler and fiber types, and environmental conditions. Based on the feed-forward back propagation algorithm, systematic analyses were performed to improve the accuracy of prediction and determine the most appropriate network topology. Training, testing, and validation results indicated that residual compressive strength of self-compacted concrete, exposed to high temperatures and relative humidity levels, could be estimated precisely with the suggested model. As illustrated by statistical indices, the reliability between experimental and predicted results was excellent. With new ingredients and different environmental conditions, the proposed model is an efficient approach to estimate the residual compressive strength of self-compacted concrete as a substitute for sophisticated laboratory procedures.

• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 2003.07a
• /
• pp.58-62
• /
• 2003

Silicon carbide (SiC) was etched in a $NF_3/CH_4$ inductively coupled plasma. The etch process was modeled by using a neural network called generalized regression neural network (GRNN). For modeling, the process was characterized by a $2^4$ full factorial experiment with one center point. To test model appropriateness, additional test data of 16 experiments were conducted. Particularly, the GRNN predictive capability was drastically improved by a genetic algorithm (GA). This was demonstrated by an improvement of more than 80% compared to a conventionally obtained model. Predicted model behaviors were highly consistent with actual measurements. From the optimized model, several plots were generated to examine etch rate variation under various plasma conditions. Unlike the typical behavior, the etch rate variation was quite different depending on the bias power Under lower bias powers, the source power effect was strongly dependent on induced dc bias. The etch rate was strongly correated to the do bias induced by the gas ratio. Particularly, the etch rate variation with the bias power at different gas ratio seemed to be limited by the etchant supply.

• Journal of the Korean Society for Precision Engineering
• /
• v.15 no.3
• /
• pp.50-60
• /
• 1998

This paper is concerned with the design of two residual control charts for real-time monitoring of the continuous flow processes. Two different control charts are designed under the situation that observations are correlated each other. Kalman-Filter based model estimation is employed when the process model is known. A black-box approach, based on Back-Propagation Neural Network, is also applied for the design of control chart when there is no prior information of process model. Performance of the designed control charts and traditional control charts is evaluated. Average run length(ARL) is adopted as a criterion for comparison. Experimental results show that the designed control chart using the Neural Network's modeling has shorter ARL than that of the other control charts when process mean is shifted. This means that the designed control chart detects the out-of-control state of the process faster than the others. The designed control chart using the Kalman-Filter based model estimation also has better performance than traditional control chart when process is out-of-control state.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.147-147
• /
• 2018

In this article, we use an open source software library: TensorFlow, developed for the purposes of conducting very complex machine learning and deep neural network applications. However, the system is general enough to be applicable in a wide variety of other domains as well. The proposed model based on a deep neural network model, LSTM (Long Short-Term Memory) to predict the river water level at Okcheon Station of the Guem River without utilization of rainfall - forecast information. For LSTM modeling, the input data is hourly water level data for 15 years from 2002 to 2016 at 4 stations includes 3 upstream stations (Sutong, Hotan, and Songcheon) and the forecasting-target station (Okcheon). The data are subdivided into three purposes: a training data set, a testing data set and a validation data set. The model was formulated to predict Okcheon Station water level for many cases from 3 hours to 12 hours of lead time. Although the model does not require many input data such as climate, geography, land-use for rainfall-runoff simulation, the prediction is very stable and reliable up to 9 hours of lead time with the Nash - Sutcliffe efficiency (NSE) is higher than 0.90 and the root mean square error (RMSE) is lower than 12cm. The result indicated that the method is able to produce the river water level time series and be applicable to the practical flood forecasting instead of hydrologic modeling approaches.

• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 2003.07a
• /
• pp.252-255
• /
• 2003

The pulsed laser deposition process modeling is investigated using neural networks based on radial basis function networks and multi-layer perceptron. Two input factors are examined with respect to the PL intensity. In order to minimize the joint confidence region of fabrication process with varying the conditions, D-optimal experimental design technique is performed and photoluminescence intensity is characterized by neural networks. The statistical results were then used to verify the fitness of the nonlinear process model. Based on the results, this modeling methodology can be optimized process conditions for pulsed laser deposition process.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.19 no.7
• /
• pp.624-630
• /
• 2006

ZnO deposition parameters are not independent and have a nonlinear and complex property. To propose a method that could verify and predict the relations of process variables, neural network was used. At first, ZnO thin films were deposited by using RF magnetron sputtering process with various conditions. Si, GaAs, and Glass were used as substrates. The temperature, work pressure, and RF power of the substrate were $50\sim500^{\circ}C$, 15 mTorr, and $180\sim210W$, respectively : the purity of the target was ZnO 4 N. Structural properties of ZnO thin films were estimated by using XRD (0002) peak intensity. The structure of neural network was a form of 4-7-1 that have one hidden layer. In training a network, learning rate and momentum were selected as 0.2, 0.6 respectively. A backpropagation neural network were performed with XRD (0002) peak data. After training a network, the temperature of substrate was evaluated as the most important parameter by sensitivity analysis and response surface. As a result, neural network could capture nonlinear and complex relationships between process parameters and predict structural properties of ZnO thin films with a limited set of experiments.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.11 no.3
• /
• pp.215-222
• /
• 2001

Neural network based models of semiconductor manufacturing processes have been shown to offer advantages in both accuracy and generalization over traditional methods. However, model development is often complicated by the fact that back-propagation neural networks contain several adjustable parameters whose optimal values unknown during training. These include learning rate, momentum, training tolerance, and the number of hidden layer neurOnS. This paper presents an investigation of the use of genetic algorithms (GAs) to determine the optimal neural network parameters for the modeling of plasma-enhanced chemical vapor deposition (PECVD) of silicon dioxide films. To find an optimal parameter set for the neural network PECVD models, a performance index was defined and used in the GA objective function. This index was designed to account for network prediction error as well as training error, with a higher emphasis on reducing prediction error. The results of the genetic search were compared with the results of a similar search using the simplex algorithm.

• International Journal of Highway Engineering
• /
• v.19 no.4
• /
• pp.1-7
• /
• 2017

PURPOSES : This study aims to improve complex modeling of multivariable, nonlinear, and overdispersion data with an artificial neural network that has been a problem in the civil and transport sectors. METHODS: Deep learning, which is a technique employing artificial neural networks, was applied for developing a large bus fuel consumption model as a case study. Estimation characteristics and accuracy were compared with the results of conventional multiple regression modeling. RESULTS : The deep learning model remarkably improved estimation accuracy of regression modeling, from R-sq. 18.76% to 72.22%. In addition, it was very flexible in reflecting large variance and complex relationships between dependent and independent variables. CONCLUSIONS : Deep learning could be a new alternative that solves general problems inherent in conventional statistical methods and it is highly promising in planning and optimizing issues in the civil and transport sectors. Extended applications to other fields, such as pavement management, structure safety, operation of intelligent transport systems, and traffic noise estimation are highly recommended.

• Proceedings of the KIEE Conference
• /
• 2004.07d
• /
• pp.2528-2530
• /
• 2004

Neural network minimization problems are often conditioned and in this contribution way to handle this will be discussed. It is shown that a better conditioned minimization problem can be obtained if the problem is separated with respect to the linear parameters. This will increase the convergence speed of the minimization. One of the most powerful uses of neural networks is in function approximation(curve fitting)[1]. A main characteristic of this solution is that function (f) to be approximated is given not explicitly but implicitly through a set of input-output pairs, named as training set, that can be easily obtained from calibration data of the measurement system. In this context, the usage of Neural Network(NN) techniques for modeling the systems behavior can provide lower interpolation errors when compared with classical methods like polynomial interpolation. This paper solve of single-variable minimization using neural network.