• Journal of Institute of Control, Robotics and Systems
• /
• v.5 no.1
• /
• pp.95-104
• /
• 1999

In this paper we propose a fast and precise control algorithm for a mobile robot, which aims at the self-tuning control applying two multi-layered neural networks to the structure of computed torque method. Through this algorithm, the nonlinear terms of external disturbance caused by variable task environments and dynamic model errors are estimated and compensated in real time by a long term neural network which has long learning period to extract the non-linearity globally. A short term neural network which has short teaming period is also used for determining optimal gains of PID compensator in order to come over the high frequency disturbance which is not known a priori, as well as to maintain the stability. To justify the global effectiveness of this algorithm where each of the long term and short term neural networks has its own functions, simulations are peformed. This algorithm can also be utilized to come over the serious shortcoming of neural networks, i.e., inefficiency in real time.

• Water Engineering Research
• /
• v.3 no.2
• /
• pp.113-122
• /
• 2002

Accurate quantitative forecasting of rainfall for basins with a short response time is essential to predict flash floods. In this study, a Quantitative Flood Forecasting (QFF) model was developed by incorporating the evolving structure and frequency of intense weather systems and by using neural network approach. Besides using radiosonde and rainfall data, the model also used the satellite-derived characteristics of storm systems such as tropical cyclones, mesoscale convective complex systems and convective cloud clusters as input. The convective classification and tracking system (CCATS) was used to identify and quantify storm properties such as lifetime, area, eccentricity, and track. As in standard expert prediction systems, the fundamental structure of the neural network model was learned from the hydroclimatology of the relationships between weather system, rainfall production and streamflow response in the study area. All these processes stretched leadtime up to 18 hours. The QFF model will be applied to the mid-Atlantic region of United States in a forthcoming paper.

• Journal of Information Technology Applications and Management
• /
• v.15 no.4
• /
• pp.21-36
• /
• 2008

As the great disturbance from abusing credit cards in Korea becomes stabilized, credit card companies need to interpret credit-card delinquents classification models from the viewpoint of profit. However, hit ratio which has been used as a measure of goodness of classification models just tells us how much correctly they classified rather than how much profits can be obtained as a result of using classification models. In this research, we tried to develop a new utility-based measure from the viewpoint of profit and then used this new measure to analyze two classification models(Neural Networks and Decision Tree models). We found that the hit ratio of neural model is higher than that of decision tree model, but the utility value of decision tree model is higher than that of neural model. This experiment shows the importance of utility based measure for credit-card delinquents classification models. We expect this new measure will contribute to increasing profits of credit card companies.

• The KIPS Transactions:PartB
• /
• v.14B no.7
• /
• pp.531-540
• /
• 2007

In this research, we proposed a two-phase hybrid stock price forecasting model with cointegration tests and artificial neural networks. Using not only the related stocks to the target stock but also the past information as input features in neural networks, the new model showed an improved performance in forecasting than that of the usual neural networks. Firstly in order to extract stocks which have long run relationships with the target stock, we made use of Johansen's cointegration test. In stock market, some stocks are apt to vary similarly and these phenomenon can be very informative to forecast the target stock. Johansen's cointegration test provides whether variables are related and whether the relationship is statistically significant. Secondly, we learned the model which includes lagged variables of the target and related stocks in addition to other characteristics of them. Although former research usually did not incorporate those variables, it is well known that most economic time series data are depend on its past value. Also, it is common in econometric literatures to consider lagged values as dependent variables. We implemented a price direction forecasting system for KOSPI index to examine the performance of the proposed model. As the result, our model had 11.29% higher forecasting accuracy on average than the model learned without cointegration test and also showed 10.59% higher on average than the model which randomly selected stocks to make the size of the feature set same as that of the proposed model.

• Database Research
• /
• v.34 no.3
• /
• pp.3-13
• /
• 2018

Text classification is one of the text mining technologies that classifies a given textual document into its appropriate categories and is used in various fields such as spam email detection, news classification, question answering, emotional analysis, and chat bot. In general, the text classification system utilizes machine learning algorithms, and among a number of algorithms, naïve Bayes and support vector machine, which are suitable for text data, are known to have reasonable performance. Recently, with the development of deep learning technology, several researches on applying deep neural networks such as recurrent neural networks (RNN) and convolutional neural networks (CNN) have been introduced to improve the performance of text classification system. However, the current text classification techniques have not yet reached the perfect level of text classification. This paper focuses on the fact that the text data is expressed as a vector only with the word dimensions, which impairs the semantic information inherent in the text, and proposes a neural network architecture based upon the semantic tensor space model.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.1505-1509
• /
• 2005

This paper introduces a new architecture of genetically optimized self-organizing fuzzy polynomial neural networks by means of information granulation. The conventional SOFPNNs developed so far are based on mechanisms of self-organization and evolutionary optimization. The augmented genetically optimized SOFPNN using Information Granulation (namely IG_gSOFPNN) results in a structurally and parametrically optimized model and comes with a higher level of flexibility in comparison to the one we encounter in the conventional FPNN. With the aid of the information granulation, we determine the initial location (apexes) of membership functions and initial values of polynomial function being used in the premised and consequence part of the fuzzy rules respectively. The GA-based design procedure being applied at each layer of genetically optimized self-organizing fuzzy polynomial neural networks leads to the selection of preferred nodes with specific local characteristics (such as the number of input variables, the order of the polynomial, a collection of the specific subset of input variables, and the number of membership function) available within the network. To evaluate the performance of the IG_gSOFPNN, the model is experimented with using gas furnace process data. A comparative analysis shows that the proposed IG_gSOFPNN is model with higher accuracy as well as more superb predictive capability than intelligent models presented previously.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.32 no.2
• /
• pp.132-139
• /
• 2009

Yield is a very important measure that can expresses simply for productivity and performance of company. So, yield is used widely in many industries nowadays. With the development of the information technology and online based real-time process monitoring technology, many industries operate the production lines that are developed into automation system. In these production lines, the product structures are very complexity and variety. So, there are many multi-variate processes that need to be monitored with many quality characteristics and associated process variables at the same time. These situations have made it possible to obtain super-large manufacturing process data sets. However, there are many difficulties with finding the cause of process variation or useful information in the high capacity database. In order to solve this problem, neural networks technique is a favorite technique that predicts the yield of process for process control. This paper uses a neural networks technique for improvement and maintenance of yield in manufacturing process. The purpose of this paper is to model the prediction of a sub process that has much effect to improve yields in total manufacturing process and the prediction of adjustment values of this sub process. These informations feedback into the process and the process is adjusted. Also, we show that the proposed model is useful to the manufacturing process through the case study.

• Korean Journal of Agricultural Science
• /
• v.50 no.3
• /
• pp.399-406
• /
• 2023

In this study, we propose a ripe tomato bunch localization method based on convolutional neural networks, to be applied in robotic harvesting systems. Tomato images were obtained from a smart greenhouse at the Rural Development Administration (RDA). The sample images for training were extracted based on tomato maturity and resized to 128 × 128 pixels for use in the classification model. The model was constructed based on four-layer convolutional neural networks, and the classes were determined based on stage of maturity, using a Softmax classifier. The localization of the ripe tomato bunch region was indicated on a class activation map. The class activation map could show the approximate location of the tomato bunch but tends to present a local part or a large part of the ripe tomato bunch region, which could lead to poor performance. Therefore, we suggest a recursive method to improve the performance of the model. The classification results indicated that the accuracy, precision, recall, and F1-score were 0.98, 0.87, 0.98, and 0.92, respectively. The localization performance was 0.52, estimated by the Intersection over Union (IoU), and through input recursion, the IoU was improved by 13%. Based on the results, the proposed localization of the ripe tomato bunch area can be incorporated in robotic harvesting systems to establish the optimal harvesting paths.

• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.48 no.11
• /
• pp.1464-1471
• /
• 1999

This paper introduces complex-valued competitive learning neural network for spatio-temporal pattern recognition. There have been quite a few neural networks for spatio-temporal pattern recognition. Among them, recurrent neural network, TDNN, and avalanche model are acknowledged as standard neural network paradigms for spatio-temporal pattern recognition. Recurrent neural network has complicated learning rules and does not guarantee convergence to global minima. TDNN requires too many neurons, and can not be regarded to deal with spatio-temporal pattern basically. Grossberg's avalanche model is not able to distinguish long patterns, and has to be indicated which layer is to be used in learning. In order to remedy drawbacks of the above networks, unsupervised competitive learning using complex umber is proposed. Suggested neural network also features simultaneous recognition, time-shift invariant recognition, stable categorizing, and learning rate modulation. The network is evaluated by computer simulation with randomly generated patterns.

• Proceedings of the KIEE Conference
• /
• 1997.11a
• /
• pp.675-678
• /
• 1997

In this research we continue the study of nuclear power plant steam generator's intelligent modeling. This model represents the input-output behavior and is a preliminary stage for intelligent control. Among many intelligent models available, we study neural network models that have been proven as universal function approximators. We select multilayer perceptrons, circular backpropagation networks, piecewise linearly trained networks and recurrent neural networks as the candidates for the steam generator's intelligent models. We take the input-output pairs from steam generator's reference model and train the neural network models. We validate trained neural network models as intelligent models of steam generator.