• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.200-200
• /
• 2000

Highly nonlinear dynamical systems are easily identified using neural networks. When disturbances are included in the learning data set Int system modeling, modeling process will be poorly performed. Since the radial basis functions in the radial basis function network(RBFN) are centered at the points specified by the weights, RBF networks are robust for approximating the process including the narrow-band disturbances deviating significantly from the regular signals. To exclude(filter) these disturbances, a robust algorithm for system identification, based on the RBFN, is proposed. The performance of system identification excluding disturbances is investigated and compared with the one including disturbances.

• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 1999.10a
• /
• pp.323-335
• /
• 1999

In this paper, we propose a new hybrid model of artificial neural networks(ANNs) and genetic algorithm (GA) to optimal feature transformation and feature weighting. Previous research proposed several variants of hybrid ANNs and GA models including feature weighting, feature subset selection and network structure optimization. Among the vast majority of these studies, however, ANNs did not learn the patterns of data well, because they employed GA for simple use. In this study, we incorporate GA in a simultaneous manner to improve the learning and generalization ability of ANNs. In this study, GA plays role to optimize feature weighting and feature transformation simultaneously. Globally optimized feature weighting overcome the well-known limitations of gradient descent algorithm and globally optimized feature transformation also reduce the dimensionality of the feature space and eliminate irrelevant factors in modeling ANNs. By this procedure, we can improve the performance and enhance the generalisability of ANNs.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.58 no.6
• /
• pp.1223-1229
• /
• 2009

The performance of Radial Basis Function Neural Networks depends on setting up the Radial Basis Functions over the input space which are the important design procedure of Radial Basis Function Neural Networks. The existing method to initialize the location of the radial basis functions over the input space is to use the conditional fuzzy C-means clustering. However, the researchers which are interested in the conditional fuzzy C-means clustering cannot get as good modeling performance as they expect because the conditional fuzzy C-means clustering cannot project the information which is extracted over the output space into the input space. To compensate the above mentioned drawback of the conditional fuzzy C-means clustering, we apply a fuzzy K-nearest neighbors approach to project the auxiliary information defined over the output space into the input space without lose of the information.

• Journal of the Korean Society for Precision Engineering
• /
• v.14 no.6
• /
• pp.90-98
• /
• 1997

A time-optimal control law for quick, strongly nonlinear systems has been developed and demonstrated. This procedure involves the utilzation of neural networks as state feedback controllers that learn the time-optimal control actions by means of an iterative minimization of both the final time and the final state error for the systems with constrained inputs and/or states. A neural identifier or a genetic algorithm identifier could be utilized for modeling the partially known systems and the unknown systems. The nature of neural networks as a parallel processor would circumvent the problem of "curwe of dimensionality". The control law has been demonstrated for both a torque input motor and a velocity input motor identified by a genetic algorithm called GENOCOPed GENOCOP.

• IE interfaces
• /
• v.16 no.spc
• /
• pp.28-32
• /
• 2003

In this paper, the fuzzy group method data handling-type(GMDH) neural networks and their application to the forecasting of mobile communication system are described. At present, GMDH family of modeling algorithms discovers the structure of empirical models and it gives only the way to get the most accurate identification and demand forecasts in case of noised and short input sampling. In distinction to neural networks, the results are explicit mathematical models, obtained in a relative short time. In this paper, an adaptive learning network is proposed as a kind of neuro-fuzzy GMDH. The proposed method can be reinterpreted as a multi-stage fuzzy decision rule which is called as the neuro-fuzzy GMDH. The GMDH-type neural networks have several advantages compared with conventional multi-layered GMDH models. Therefore, many types of nonlinear systems can be automatically modeled by using the neuro-fuzzy GMDH. The computer program is developed and successful applications are shown in the field of estimating problem of mobile communication with the number of factors considered.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.15 no.1
• /
• pp.7-14
• /
• 2002

In this paper, neutral networks are used to build models of oxide film etched In CHF$_3$/CF$_4$ with a magnetically enhanced reactive ion etcher(MERIE). A statistical 2$\^$4-1/ experimental design plus one center point was used to characterize relationships between process factors and etch responses. The factors that were varied include radio frequence(rf) power, pressure, CHF$_3$ and CF$_4$ flow rates. Resultant 9 experiments were used to train neural networks and trained networks were subsequently tested on its appropriateness using additionally conducted 8 experiments. A total of 17 experiments were thus conducted for this modeling. The etch responses modeled are dc bias voltage, etch rate and etch uniformity A qualitative, good agreement was obtained between predicted and observed behaviors.

• Korean Management Science Review
• /
• v.9 no.1
• /
• pp.17-30
• /
• 1992

This paper is organized as follows. Section 2 illustrates several applications of neural networks. Section 3 presents the theoretical aspects of the major neural network paradigms as well as the structure of the back -propagation network used in the study. Section 4 describes the experiment including data analysis, modeling, and the performance criteria followed by the detailed discussion of the experimental results. Future research avenues including advantages and limitations of neural network are presented in the last section.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.2
• /
• pp.251-256
• /
• 2005

Up to now, a number of models have been proposed and discussed to describe a wide range of inelastic behaviors of materials. The fetal problem of using such models is however the existence of model errors, and the problem remains inevitably as far as a material model is written explicitly. In this paper, the authors define the implicit constitutive model and propose an implicit viscoplastic constitutive model using neural networks. In their modeling, inelastic material behaviors are generalized in a state space representation and the state space form is constructed by a neural network using input output data sets. A technique to extract the input-output data from experimental data is also described. The proposed model was first generated from pseudo-experimental data created by one of the widely used constitutive models and was found to replace the model well. Then, having been tested with the actual experimental data, the proposed model resulted in a negligible amount of model errors indicating its superiority to all the existing explicit models in accuracy.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.4
• /
• pp.1035-1045
• /
• 1997

It is important to understand how we can deal with doements for the modeling of neural networks when we are unbestigating the dynamical performance and the information procoessing capabilitids.The information procewssing capabkities of model neural networks will change for different response, synaptic weights or learning rules. Using the staritical neurodyamics method, we evalute the capabikities of neural networks in order to understand the basic conept ofr parallel distributed processing. In this paper, we explain the reuslts of theoretical anaysis of the effests of the hysteretic property on the performance of wuquential associative neral networks.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.66 no.12
• /
• pp.1772-1781
• /
• 2017

In this study, the design methodology for alleviating the overfitting problem of Polynomial Neural Networks(PNN) is realized with the aid of two kinds techniques such as L2 regularization and Sum of Squared Coefficients (SSC). The PNN is widely used as a kind of mathematical modeling methods such as the identification of linear system by input/output data and the regression analysis modeling method for prediction problem. PNN is an algorithm that obtains preferred network structure by generating consecutive layers as well as nodes by using a multivariate polynomial subexpression. It has much fewer nodes and more flexible adaptability than existing neural network algorithms. However, such algorithms lead to overfitting problems due to noise sensitivity as well as excessive trainning while generation of successive network layers. To alleviate such overfitting problem and also effectively design its ensuing deep network structure, two techniques are introduced. That is we use the two techniques of both SSC(Sum of Squared Coefficients) and $L_2$ regularization for consecutive generation of each layer's nodes as well as each layer in order to construct the deep PNN structure. The technique of $L_2$ regularization is used for the minimum coefficient estimation by adding penalty term to cost function. $L_2$ regularization is a kind of representative methods of reducing the influence of noise by flattening the solution space and also lessening coefficient size. The technique for the SSC is implemented for the minimization of Sum of Squared Coefficients of polynomial instead of using the square of errors. In the sequel, the overfitting problem of the deep PNN structure is stabilized by the proposed method. This study leads to the possibility of deep network structure design as well as big data processing and also the superiority of the network performance through experiments is shown.