Neural Net Model Research Articles

시계열 자료 코스피200의 패턴분류를 위한 퍼지 서포트 벡타 기계

주식시장에서 KOSPI200지수의 상승 또는 하락으로 분류 및 예측하는 정보는 선물 및 옵션시장에서 포토폴리오를 설계할 때 의사결정을 위해 중요한 기준이 된다. 경제지표인 시계열 패턴들의 향후 추세는 가장 최근의 경제패턴에 매우 종속적이기 때문에 최근의 패턴들을 가장 우선적으로 학습해야 할 필요가 있다. 본 논문에서는 시계열분석, 신경회로망, 그리고 다양한 분야에서 각광을 받고 있는 SVM(Support Vector Machine)과 Fuzzy SVM 모형의 분류 및 예측성능을 비교하였다. 특히 학습 DB에 따라 시계열성 속성을 갖는 퍼지소속함수에 가장 적합한 차원을 제시함으로서 Fuzzy SVM이 우수함을 입증하였다. The Information of classification and estimate about KOSPI200 index`s up and down in the stock market becomes an important standard of decision-making in designing portofolio in futures and option market. Because the coming trend of time series patterns, an economic indicator, is very subordinate to the most recent economic pattern, it is necessary to study the recent patterns most preferentially. This paper compares classification and estimated performance of SVM(Support Vector Machine) and Fuzzy SVM model that are getting into the spotlight in time series analyses, neural net models and various fields. Specially, it proves that Fuzzy SVM is superior by presenting the most suitable dimension to fuzzy membership function that has time series attribute in accordance with learning Data Base.

Neural-Net Analysis of the Plasticity of ÉP-450 Ferrite–Martensite Steel with Different Alloying Element Concentrations

The results of a computational experiment with a neural-net model simulating the effect of the chemical composition on the plastic properties of EP-450 steel are presented. It is shown that computer-aided development of reactor steel with prescribed properties is, in principle, possible.

Towards an Automatic Interpretation and Knowledge Based Search of Satellite Images in Databases

This paper shows a prototype automatic interpretation system for ocean AVHRR (Advanced Very High Resolution Radiometer) satellite images. It is built on a three-level knowledge model (pixel, regional and domain semantic problem levels) and uses several connectionist computational approaches. First artificial neural net models (to the pixel level) were used for basic preprocessing tasks such as cloud masking. Next, a new connectionist technique using input vectors with nonnumerical regional marine features has also been developed and used in the identification phase. The paper shows some results of oceanic structure identification tasks (wakes, upwellings and eddies) in infrared images of the NW African coast and Canary Islands. These results illustrate a procedure for improving automatic oceanic interpretation of satellite images.

The Use of Neural Nets for Matching Fixed or Variable Geometry Compressors With Diesel Engines

A technique which uses trained neural nets to model the compressor in the context of a turbocharged diesel engine simulation is introduced. This technique replaces the usual interpolation of compressor maps with the evaluation of a smooth mathematical function. Following presentation of the methodology, the proposed neural net technique is validated against data from a truck type, 6-cylinder 14-liter diesel engine. Furthermore, with the introduction of an additional parameter, the proposed neural net can be trained to simulate an entire family of compressors. As a demonstration, a family of compressors of different sizes is represented with a single neural net model which is subsequently used for matching calculations with intercooled and nonintercooled engine configurations at different speeds. This novel approach readily allows for evaluation of various options within a wide range of possible compressor configurations prior to prototype production. It can also be used to represent the variable geometry machine regardless of the method used to vary compressor characteristics. Hence, it is a powerful design tool for selection of the best compressor for a given diesel engine system and for broader system optimization studies.

Competitive neural-net-based system for the automatic detection of oceanic mesoscalar structures on AVHRR scenes

This paper shows a prototype automatic interpretation system for Advanced Very High Resolution Radiometer satellite ocean images. It is built on a three-level knowledge model (pixel, regional, and domain semantic problem levels) and uses several connectionist computational approaches. First, artificial neural net models (to the pixel level) were used for basic preprocessing tasks such as cloud masking. Next, a new connectionist technique using input vectors with nonnumerical regional marine features has also been developed and used in the identification phase. The paper shows some results of oceanic structure identification tasks (wakes, upwellings, and eddies) in infrared images of the northwest African coast and the Canary Islands. These results illustrate a procedure for improving automatic oceanic interpretation of satellite images.

Predicting of slow noise and vibration spectra degradation in the IBR-2 pulsed neutron source using a neural network simulator

Employing a neural net model of the noise state of the IBR-2 reactor (JINR, Dubna, Russia) and a model of the vibration state of movable reflectors in the reactor we have predicted slow degradation in reactor noises. Operative diagnostics and prediction of the reactor noise behavior with time involves separating of the reflector degradation trend in power noises. We investigate two neural models. The first concerns the vibrations of the reflectors and the second is a simplified reactor noise model. The predicted results are close, in character, to the experimental data. They show that it is the movable reflectors that are mainly responsible for the degradation of power noises.

Coupling of neural network and dispersion models: a novel methodology for air pollution models

Supervised neural net models and dispersion models are two important approaches for evaluating air pollution concentrations. The authors propose the development of an integrated model, in order to optimise the performances of each methodology. The concentrations evaluated by an air pollution model are coupled with a Neural Net (NN), so as to adjust the influence of important variables on dispersion models (which may produce systematic under- or over-prediction of measured concentrations). In particular, an optimised 3-Layer Perception with error-backpropagation learning rules is used to filter the air pollution concentrations evaluated using an operative analytical model that takes account of the vertical profiles of wind and turbulent diffusivity. The results show good performances of this methodology when applied to the Kincaid dataset.

Neural Networks and Regional Science Modeling: A Survey of Techniques for Complex Spatial Analysis

This study examines the value of utilizing neural net modeling for issues relating to optimization across a network of cities in space. Neural nets are made up of many nonlinear computational elements that operate in parallel and are arranged in a manner similar to biological neural nets. Defining a neural net model involves specifying a net topology, arrangement of nodes, training or learning rules, adjustment of weights associated with connections, node characteristics, and rules of transformation from input to output. All of these are the major issues in such regional problems as labor force migration and firm location.

Control of a reactive separation process

We discuss general aspects of the control of reactive separation processes and specifically the example of the control of a semi-batch reactive distillation process. It is pointed out that more complicated controller structures, more sophisticated controller design methods, and alternative, model-based nonlinear controllers are needed compared to conventional processes. For the example process, we discuss both conventional control structures and model-based predictive control using a neural net plant model.

Comparison of Logistic Regression and Neural Net Modeling for Prediction of Prostate Cancer Pathologic Stage

Prostate cancer (PCa) pathologic staging remains a challenge for the physician using individual pretreatment variables. We have previously reported that UroScore, a logistic regression (LR)-derived algorithm, can correctly predict organ-confined (OC) disease state with >90% accuracy. This study compares statistical and neural network (NN) approaches to predict PCa stage. A subset (756 of 817) of radical prostatectomy patients was assessed: 434 with OC disease, 173 with capsular penetration (NOC-CP), and 149 with metastases (NOC-AD) in the training sample. Additionally, an OC + NOC-CP (n = 607) vs NOC-AD (n = 149) two-outcome model was prepared. Validation sets included 120 or 397 cases not used for modeling. Input variables included clinical and several quantitative biopsy pathology variables. The classification accuracies achieved with a NN with an error back-propagation architecture were compared with those of LR statistical modeling. We demonstrated >95% detection of OC PCa in three-outcome models, using both computational approaches. For training patient samples that were equally distributed for the three-outcome models, NNs gave a significantly higher overall classification accuracy than the LR approach (40% vs 96%, respectively). In the two-outcome models using either unequal or equal case distribution, the NNs had only a marginal advantage in classification accuracy over LR. The strength of a mathematics-based disease-outcome model depends on the quality of the input variables, quantity of cases, case sample input distribution, and computational methods of data processing of inputs and outputs. We identified specific advantages for NNs, especially in the prediction of multiple-outcome models, related to the ability to pre- and postprocess inputs and outputs.

A flexible brand choice model based on neural net methodology A comparison to the linear utility multinomial logit model and its latent class extension

Brand choice models as a rule have a linear (deterministic) utility function, i.e. they conceive utility as linear combination of predictors like price, sales promotion variables, brand name and other product attributes. To discover nonlinear effects on brands' utilities in a flexible way we specify deterministic utility by means of a certain type of neural net. This feedforward multilayer perceptron is able to approximate any continuous multivariate function and its derivatives with the desired level of precision. In an empirical study the neural net based choice model leads to better out-of-sample results than homogeneous and heterogeneous versions of linear utility MNL models. On the other hand the latent class variant of the linear utility MNL model attains better fit values for estimation data than the neural net model. The neural net approach implies different choice elasticities for most predictors and identifies nonlinear effects (like interaction effects, thresholds, saturation effects).

A data-centric approach to understanding the pricing of financial options

We investigate what can be learned from a purely phenomenological study of options prices without modelling assumptions. We fitted neural net (NN) models to LIFFE “ESX” European style FTSE 100 index options using daily data from 1992 to 1997. These non-parametric models reproduce the Black-Scholes (BS) analytic model in terms of fit and performance measures using just the usual five inputs (S, X, t, r, IV). We found that adding transaction costs (bid-ask spread) to these standard five parameters gives a comparable fit and performance. Tests show that the bid-ask spread can be a statistically significant explanatory variable for option prices. The difference in option prices between the models with transaction costs and those without ranges from about -3.0 to +1.5 index points, varying with maturity date. However, the difference depends on the moneyness (S/X), being greatest in-the-money. This suggests that use of a five-factor model can result in a pricing difference of up to £10 to £30 per call option contract compared with modelling under transaction costs. We found that the influence of transaction costs varied between different yearly subsets of the data. Open interest is also a significant explanatory variable, but volume is not.

Effect of harmonic relatedness on the detection of temporal asynchronies.

Speeded intonation judgments of a target chord are facilitated when the chord is preceded by a harmonically related prime chord. The present study extends harmonic priming to temporal asynchrony judgments. In both tasks, the normative target chords (consonant, synchronous) are processed more quickly and accurately after a harmonically related prime than after a harmonically unrelated prime. However, the influence of harmonic context on sensitivity (d') differs between the two tasks: d' was higher in the related context for intonation judgments but was higher in the unrelated context for asynchrony judgments. A neural net model of tonal knowledge activation provides an explanatory framework for both the facilitation in the related contexts and the sensitivity differences between the tasks.

A HYBRID APPROACH TO KNOWLEDGE ACQUISITION USING NEURAL PETRI NETS AND DS THEORY

Knowledge acquisition from multiple experts and its refinement are important issues in knowledge management of an expert system. The paper presents a novel approach to handling the above problems by combining the synergistic behavior of neural Petri nets and the Dempster–Shafer theory. The Dempster–Shafer theory has been employed here to reduce the scope of uncertainty in the supplied noisy input instances and the inferences generated therefrom by the multiple experts. The noise-free training instances thus obtained are subsequently used to train the neural Petri net model for refining the parameters of its knowledge base. A comparison of the performance of the proposed training algorithm with the classical back-propagation algorithm has also been presented in the paper.

Model-following Control of Nonlinear Servo Systems Using Neural Networks

The most useful property of neural networks for system identification and control is their ability to approximate arbitrary nonlinear mappings. In this paper, a new neural network model-following control method is proposed for nonlinear servo systems. First, three-layer artificial neural net works are trained to describe the forward and in verse dynamic characteristics of one type of speci fied nonlinear servo systems by using the backpropagation algorithm. Then the neural net work inverse model is used as a feedforward con troller for model-following control of the nonlin ear servo systems. Computer simulation results obtained from MATLAB verify the applicability of neural network identification and control for nonlinear servo systems.

A nonlinear industrial model predictive controller using integrated PLS and neural net state-space model

Model predictive control (MPC) technology has been well developed and successfully applied in the refinery and petrochemical process industries over the last 20 years. Recent development has been focused on nonlinear MPC and robust MPC technologies because new challenges have been encountered in the polymer and chemical industries where many processes show strong nonlinearity and uncertainty. This paper presents a nonlinear industrial model predictive controller, recently developed by Aspen Technology Inc. This MPC controller uses a nonlinear, state-space, integrated partial least-squares (PLS) and neural net model (Zhao, Guiver and Sentoni, American control conference, Philadelphia, PA, USA, 1998), and a multi-step, constrained, Newton-type optimization algorithm (Oliveira and Biegler, Automatica, 31 (2) (1995) 281–286). It results in a robust and cost-effective industrial nonlinear MPC controller. A pH reactor example and a successful industrial application in NO x emission control of a power plant are presented to demonstrate the capability of this controller.

Reorientation in a competitive environment: An analysis of strategic change

Attraction models are very popular in marketing research for studying the effects of marketing instruments on market shares. However, so far the marketing literature only considers attraction models with certain functional forms that exclude threshold or saturation effects on attraction values. We can achieve greater flexibility by using the neural net based approach introduced here. This approach assesses brands’ attraction values by means of a perception with one hidden layer. The approach uses log-ratio transformed market shares as dependent variables. Stochastic gradient descent followed by a quasi-Newton method estimates parameters. For store-level data, neural net models perform better and imply a price response that is qualitatively different from the well-known multinomial logit attraction model. Price elasticities of neural net attraction models also lead to specific managerial implications in terms of optimal prices.

MULTIPLE FEATURE ANALYSIS FOR MACHINE VISION GRADING OF FLORIDA CITRUS

Both blemish and physical attributes were acquired on commercially graded Florida grapefruit, orange, andtangerine varieties. Using equal numbers of acceptable and rejected fruit, various neural network classification strategieswere applied to blemishrelated features and blemish plus physical features. The blemish plus physical feature neural netmodels were the most successful, yielding overall correct classification levels of 98.5% for grapefruit and orange and 98.3%for tangerine. No significant difference was found between the neural net models of standard backpropagation, jump step,or variable transfer functions for the hidden layer.

An artificial neural net attraction model (ANNAM) to analyze market share effects of marketing instruments

Attraction models are very popular in marketing research for studying the effects of marketing instruments on market shares. However, so far the marketing literature only considers attraction models with certain functional forms that exclude threshold or saturation effects on attraction values. We can achieve greater flexibility by using the neural net based approach introduced here. This approach assesses brands’ attraction values by means of a perception with one hidden layer. The approach uses log-ratio transformed market shares as dependent variables. Stochastic gradient descent followed by a quasi-Newton method estimates parameters. For store-level data, neural net models perform better and imply a price response that is qualitatively different from the well-known multinomial logit attraction model. Price elasticities of neural net attraction models also lead to specific managerial implications in terms of optimal prices.

Modelling of the river flowrate: the influence of the training set selection

A study of the influence of the training set selection, the modelling technique, and the number of objects in the training set was performed on a data set of 2 years' daily measurements of atmospheric precipitation and river flowrates. Twenty-five different data sets were prepared by the following selection methods: random selection (RND), Kohonen neural network selection, and Kennard–Stone selection (K–S). On these data sets, 125 models (regressions) were generated using the following five methods: multiple linear regression (MLR), partial least squares regression (PLSR) and two feed-forward neural networks with the error back-propagation and Levenberg–Marquardt learning algorithm (LM). The models were tested using a single set separated from the rest of the data. Additionally, a bottleneck neural network with back-propagation learning was tested, since it combines both modelling and mapping capabilities. The results are discussed and two examples of model use are presented. The best method for sample division into the training/monitoring sets was the Kohonen map (KOH) division. The best model obtained among those generated yields the test set RMS error of 0.0845 and was achieved by L–M neural net model on 400 randomly selected training objects.