Conventional Feedforward Neural Networks Research Articles

Force control of a magnetorheological damper using an elementary hysteresis model-based feedforward neural network

An inverse controller is proposed for a magnetorheological (MR) damper that consists of a hysteresis model and a voltage controller. The force characteristics of the MR damper caused by excitation signals are represented by a feedforward neural network (FNN) with an elementary hysteresis model (EHM). The voltage controller is constructed using another FNN to calculate a suitable input signal that will allow the MR damper to produce the desired damping force. The performance of the proposed EHM-based FNN controller is experimentally compared to existing control methodologies, such as clipped-optimal control, signum function control, conventional FNN, and recurrent neural network with displacement or velocity inputs. The results show that the proposed controller, which does not require force feedback to implement, provides excellent accuracy, fast response time, and lower energy consumption.

Radial basis function neural networks applied to efficient QRST cancellation in atrial fibrillation

The most extended noninvasive technique for medical diagnosis and analysis of atrial fibrillation (AF) relies on the surface elctrocardiogram (ECG). In order to take optimal profit of the ECG in the study of AF, it is mandatory to separate the atrial activity (AA) from other cardioelectric signals. Traditionally, template matching and subtraction (TMS) has been the most widely used technique for single-lead ECGs, whereas multi-lead ECGs have been addressed through statistical signal processing techniques, like independent component analysis. In this contribution, a new QRST cancellation method based on a radial basis function (RBF) neural network is proposed. The system is able to provide efficient QRST cancellation and can be applied both to single and multi-lead ECG recordings. The learning algorithm used for training the RBF makes use of a special class of network, known as cosine RBF, by updating selected adjustable parameters to minimize the class-conditional variances at the outputs of the network. The experiments verify that RBFs trained by the proposed learning algorithm are capable of reducing the QRST complex dramatically, a property that is not shared by other methods and conventional feed-forward neural networks. Average Results (mean ± std) for the RBF method in cross-correlation (CC) between original and estimated AA are CC=0.95±0.038 being the mean square error (MSE) for the same signals, MSE=0.311±0.078. Regarding spectral parameters, the dominant amplitude (DA) and the mean power spectral (MP) were DA=1.15±0.18 and MP=0.31±0.07, respectively. In contrast, traditional TMS-based methods yielded, for the best case, CC=0.864±0.041, MSE=0.577±0.097, DA=0.84±0.25 and MP=0.24±0.07. The results prove that the RBF based method is able to obtain a remarkable reduction of ventricular activity and a very accurate preservation of the AA, thus providing high quality dissociation between atrial and ventricular activities in AF recordings.

Wavelet neural network model for reservoir inflow prediction

In this study, a Wavelet Neural Network (WNN) model is proposed for monthly reservoir inflow prediction by combining the Discrete Wavelet Transform (DWT) and Levenberg-Marquardt optimization algorithm-based Feed Forward Neural Networks (FFNN). The study area covers the basin of Kemer Dam which is located in the Aegean region of Turkey. Monthly meteorological data were decomposed into wavelet sub-time series by DWT. Ineffective sub-time series have been eliminated by using all possible regression method and evaluating the Mallows’ Cp coefficients to prevent collinearity. Then, effective sub-time series components have been used as the new inputs of neural networks. DWT has been also integrated with multiple linear regressions (WREG) within the study. The results of Wavelet Neural Network (WNN) model and WREG have been compared with conventional Feed Forward Neural Networks (FFNN) and multiple linear regression (REG) models. When the statistical-based criteria are examined, it has been observed that the DWT method has increased the performances of feed forward neural networks and regression methods. The results determined in the study indicate that the WNN is a successful tool to model the monthly inflow series of dam and can give good prediction performances than other methods.

The Prediction of Wax Precipitation by Neural Network and Genetic Algorithm and Comparison With a Multisolid Model in Crude Oil Systems

This article introduce a new implementation of the neural network and genetic programming neural network technology in petroleum engineering. An intelligent framework is developed for calculating the amount of wax precipitation in petroleum mixtures over a wide temperature range. Theoretical results and practical experience indicate that feedforward networks can approximate a wide class of function relationships very well. In this work, a conventional feedforward multilayer neural network and genetic programming neural network (GPNN) approach have been proposed to predict the amount of wax precipitation. The introduced model can predict wax precipitation through neural network and genetic algorithmic techniques. The accuracy of the method is evaluated by predicting the amount of wax precipitation of various reservoir fluids not used in the development of the models. Furthermore, the performance of the model is compared with the performance of multisolid model for wax precipitation prediction and experimental data. Results of this comparison show that the proposed method is superior, both in accuracy and generality, over the other models.

Least squares support vector machine for modeling daily reference evapotranspiration

The accuracy of a least square support vector machine (LSSVM) in modeling of reference evapotranspiration (ET0) was examined in this study. The daily weather data, solar radiation, air temperature, relative humidity and wind speed of two stations, Glendale and Oxnard, in southern district of California, were used as inputs to the LSSVM models to estimate ET0 obtained using the FAO-56 Penman–Monteith equation. In the first part of the study, LSSVM estimates were compared with those of the following empirical models: Priestley–Taylor, Hargreaves and Ritchie methods. The comparison results indicated that the LSSVM performed better than the empirical models. In the second part of the study, the LSSVM results were compared with those of the conventional feed-forward artificial neural networks (ANN). It was found that the LSSVM models were superior to the ANN in modeling ET0 process.

Extreme Learning Machine for Regression and Multiclass Classification

Due to the simplicity of their implementations, least square support vector machine (LS-SVM) and proximal support vector machine (PSVM) have been widely used in binary classification applications. The conventional LS-SVM and PSVM cannot be used in regression and multiclass classification applications directly, although variants of LS-SVM and PSVM have been proposed to handle such cases. This paper shows that both LS-SVM and PSVM can be simplified further and a unified learning framework of LS-SVM, PSVM, and other regularization algorithms referred to extreme learning machine (ELM) can be built. ELM works for the "generalized" single-hidden-layer feedforward networks (SLFNs), but the hidden layer (or called feature mapping) in ELM need not be tuned. Such SLFNs include but are not limited to SVM, polynomial network, and the conventional feedforward neural networks. This paper shows the following: 1) ELM provides a unified learning platform with a widespread type of feature mappings and can be applied in regression and multiclass classification applications directly; 2) from the optimization method point of view, ELM has milder optimization constraints compared to LS-SVM and PSVM; 3) in theory, compared to ELM, LS-SVM and PSVM achieve suboptimal solutions and require higher computational complexity; and 4) in theory, ELM can approximate any target continuous function and classify any disjoint regions. As verified by the simulation results, ELM tends to have better scalability and achieve similar (for regression and binary class cases) or much better (for multiclass cases) generalization performance at much faster learning speed (up to thousands times) than traditional SVM and LS-SVM.

SUPPORT VECTOR REGRESSION AND FUNCTIONAL NETWORKS FOR VISCOSITY AND GAS/OIL RATIO CURVES ESTIMATION

In oil and gas industry, prior prediction of certain properties is needed ahead of exploration and facility design. Viscosity and gas/oil ratio (GOR) are among those properties described through curves with their values varying over a specific range of reservoir pressures. However, the usual single point prediction approach could result into curves that are inconsistent, exhibiting scattered behavior as compared to the real curves. Support Vector Regressors and Functional Networks are explored in this paper to solve this problem. Inputs into the developed models include hydrocarbon and non-hydrocarbon crude oil compositions and other strongly correlating reservoir parameters. Graphical plots and statistical error measures, including root mean square error and average absolute percent relative error, have been used to evaluate the performance of the models. A comparative study is performed between the two techniques and with the conventional feed forward artificial neural networks. Most importantly, the predicted curves are consistent with the shapes of the physical curves of the mentioned oil properties, preserving the need of such curves for interpolation and ensuring conformity of the predicted curves with the conventional properties.

Performance of fuzzy logic-based slope tuning of neural equaliser for digital communication channel

Adaptive equalisation in digital communication systems is a process of compensating the disruptive effects caused mainly by intersymbol interference in a band-limited channel and plays a vital role for enabling higher data rate in modern digital communication system. Designing efficient equalisers having low structural complexity and faster learning algorithms is also an area of much research interest in the present scenario. This paper presents a novel technique of improving the performance of conventional multilayer perceptron (MLP)-based decision feedback equaliser (DFE) of reduced structural complexity by adapting the slope of the sigmoidal activation function using fuzzy logic control technique. The adaptation of the slope parameter increases the degrees of freedom in the weight space of the conventional feedforward neural network (CFNN) configuration. Application of this technique provides faster learning with less training samples and significant performance gain. This research work also proposes adaptive channel equalisation techniques on recurrent neural network framework. Exhaustive simulation studies carried out prove that by replacing the conventional sigmoid activation functions in each of the processing nodes of recurrent neural network with multilevel sigmoid activation functions, the bit error rate performance has significantly improved. Further slopes of different levels of the multilevel sigmoid have been adapted using fuzzy logic control concept. Simulation results considering standard channel models show faster learning with less number of training samples and performance level comparable to the their conventional counterparts. Also, there is scope for parallel implementation of slope adaptation technique in real-time implementation, which saves the computational time.

Any-Place Forecasting Method of Nationwide Time-Series Wind Speed Using Classified Forecast Models based on Wind Conditions

Most of the methods forecasting the wind speed proposed so far are not universal and available only for specific area. In the present study, forecasting method of a time-series wind speed 24-hours ahead at any location in Japan was proposed, which was based on classified forecast models by wind conditions (called FCCM; Forecasting Method using Classified forecast Model). The model was structured by the radial basis function network (RBFN) and the performance was compared to the conventional feed forward neural network (FFNN). The procedure of the FCCM was as follows; (1) All the points where wind speed had been measured by the Japan Meteorological Agency were divided into the groups depending on their wind conditions (mean wind speed and standard deviation of wind speed). (2) RBFN was learned and the network model was optimized to forecast wind speed in each group. (3) If the wind speed was not measured in the forecast point, it was estimated by using measured data around there. (4) Then, the time-series wind speed of the forecast point was obtained by the network model for the group. The wind speeds at 6 points in Japan where the wind condition were different were forecasted by FCCM. As a result, the mean absolute error was about from 1.0 to 1.5m/s. Although forecast error of RBFN and that of FFNN were similar, RBFN was able to learn fast than FFNN.

Comparison of performance prediction of solar water heaters between Artificial Neural Networks and conventional correlations

The aim of this study was to develop a predictive method for heat transfer coefficients in solar water heaters and their performance evaluation of such heaters with different materials used as heat collectors. Two approaches have been used: conventional method and an Artificial Neural Network (ANN) to predict the performance of solar water heaters and to compare these two approaches. This performance is measured in terms of outlet temperature by using a set of conventional feed forward multi-layer neural networks. The actual experimental data which were used as our network's input gathered from published literature (for polypropylene tubes) and from the experiments carried out recently using copper tubes. The results of this study showed that ANN approach can give better approximation than the traditional theoretical correlations which was obtained by linear regression analysis.

High‐sensitive neural network ammonia sensor based on shear horizontal surface acoustic wave devices

AbstractIn this paper, a shear horizontal surface acoustic wave devices coated with L‐glutamic acid hydrochloride were applied as ammonia sensors. This sensor has shown high sensitivity and fast responses to ppb‐level ammonia. The frequency shift linearly increased as the ammonia concentration increased from 40 to 400 ppb in dry environment. In the humid environment, the frequency shift gradually decreased with ammonia concentration increasing. In order to precisely estimate the ammonia in humid environment, two different neural models, the conventional feedforward neural network and quantum neural network, were used as the identifier and their performances were reported and compared. Copyright © 2008 John Wiley & Sons, Ltd.

Automated dental recognition by wavelet descriptors in CT multi-slices data

Teeth are one of the most important anatomical components in ergonomic appearance of the face and crucial in craniofacial surgeries and dentistry. Besides, teeth have specific roles in forensic to identify death persons. In this paper, we propose a hybrid technique to classify teeth in computed tomography (CT) dental images. Our method consists of three steps: segmentation, feature extraction and classification. We segment the teeth by a hybrid approach including anatomical-based histogram thresholding, panaromic resampling and level-set techniques. In the second step, for each segmented tooth, we calculate the 2D discrete wavelet transform. We then determine 1D inverse discrete wavelet transform by the same mother wavelet, calculate energy of decomposition and approximation coefficients. Next, we utilize these coefficients as feature vectors for classification in the third step. Teeth classification is performed by a conventional feed forward neural network. The proposed method is evaluated in the presence of 180 teeth. Experimental results reveal that the technique is effective to automatically classify teeth, on an average, in more than 95% of the cases in both jaws. Our method is independent of anatomical information such as the sequence and locality of the teeth in jaws. The techniques are applicable to both forensic and dentistry investigations.

Is bias dispensable for fuzzy neural networks?

It is well known that a conventional feedforward neural network always has bias terms, which is necessary for it to solve classification or approximation problems. But for fuzzy neural networks (FNNs), it seems not quite clear yet whether the bias is dispensable or not. Some authors introduce the bias, while the others do not. This note tries to partly answer this question for two simple building blocks of FNNs: fuzzy perceptron and max–min FNN. It is shown that the bias is basically dispensable for fuzzy perceptrons (max–min FNNs plus a sign function), which are usually used for classification. On the other hand, the bias is dispensable for max–min FNNs, which are used for both approximation and classification, if and only if a special condition is valid. But this special condition is generally not valid, or not easy to justify in practice. So the bias is generally indispensable for max–min FNNs.

Semi-Parametric Learning for Classification of Pitting Corrosion Detected by Acoustic Emission

This paper presents a Non-Destructive Testing (NDT) technique, Acoustic Emission (AE) to classify pitting corrosion severity in austenitic stainless steel 304 (SS304). The corrosion severity is graded roughly into five levels based on the depth of corrosion. A number of timedomain AE parameters were extracted and used as features in our classification methods. In this work, we present practical classification techniques based on Bayesian Statistical Decision Theory, namely Maximum A Posteriori (MAP) and Maximum Likelihood (ML) classifiers. Mixture of Gaussian distributions is used as the class-conditional probability density function for the classifiers. The mixture model has several appealing attributes such as the ability to model any probability density function (pdf) with any precision and the efficiency of parameter-estimation algorithm. However, the model still suffers from model-order-selection and initialization problems which greatly limit its applications. In this work, we introduced a semi-parametric scheme for learning the mixture model which can solve the mentioned difficulties. The method was compared with conventional Feed-Forward Neural Network (FFNN) and Probabilistic Neural Network (PNN) to evaluate its performance. We found that our proposed methods gave much lower classificationerror rate and also far smaller variance of the classifiers.

Training Reformulated Radial Basis Function Neural Networks Capable of Identifying Uncertainty in Data Classification

This paper introduces a learning algorithm that can be used for training reformulated radial basis function neural networks (RBFNNs) capable of identifying uncertainty in data classification. This learning algorithm trains a special class of reformulated RBFNNs, known as cosine RBFNNs, by updating selected adjustable parameters to minimize the class-conditional variances at the outputs of their radial basis functions (RBFs). The experiments verify that quantum neural networks (QNNs) and cosine RBFNNs trained by the proposed learning algorithm are capable of identifying uncertainty in data classification, a property that is not shared by cosine RBFNNs trained by the original learning algorithm and conventional feed-forward neural networks (FFNNs). Finally, this study leads to a simple classification strategy that can be used to improve the classification accuracy of QNNs and cosine RBFNNs by rejecting ambiguous feature vectors based on their responses.

Automated Detection of Videotaped Neonatal Seizures of Epileptic Origin

This study aimed at the development of a seizure-detection system by training neural networks with quantitative motion information extracted from short video segments of neonatal seizures of the myoclonic and focal clonic types and random infant movements. The motion of the infants' body parts was quantified by temporal motion-strength signals extracted from video segments by motion-segmentation methods based on optical flow computation. The area of each frame occupied by the infants' moving body parts was segmented by clustering the motion parameters obtained by fitting an affine model to the pixel velocities. The motion of the infants' body parts also was quantified by temporal motion-trajectory signals extracted from video recordings by robust motion trackers based on block-motion models. These motion trackers were developed to adjust autonomously to illumination and contrast changes that may occur during the video-frame sequence. Video segments were represented by quantitative features obtained by analyzing motion-strength and motion-trajectory signals in both the time and frequency domains. Seizure recognition was performed by conventional feed-forward neural networks, quantum neural networks, and cosine radial basis function neural networks, which were trained to detect neonatal seizures of the myoclonic and focal clonic types and to distinguish them from random infant movements. The computational tools and procedures developed for automated seizure detection were evaluated on a set of 240 video segments of 54 patients exhibiting myoclonic seizures (80 segments), focal clonic seizures (80 segments), and random infant movements (80 segments). Regardless of the decision scheme used for interpreting the responses of the trained neural networks, all the neural network models exhibited sensitivity and specificity>90%. For one of the decision schemes proposed for interpreting the responses of the trained neural networks, the majority of the trained neural-network models exhibited sensitivity>90% and specificity>95%. In particular, cosine radial basis function neural networks achieved the performance targets of this phase of the project (i.e., sensitivity>95% and specificity>95%). The best among the motion segmentation and tracking methods developed in this study produced quantitative features that constitute a reliable basis for detecting neonatal seizures. The performance targets of this phase of the project were achieved by combining the quantitative features obtained by analyzing motion-strength signals with those produced by analyzing motion-trajectory signals. The computational procedures and tools developed in this study to perform off-line analysis of short video segments will be used in the next phase of this project, which involves the integration of these procedures and tools into a system that can process and analyze long video recordings of infants monitored for seizures in real time.

Detection of Pseudosinusoidal Epileptic Seizure Segments in the Neonatal EEG by Cascading a Rule-Based Algorithm With a Neural Network

This paper presents an approach to detect epileptic seizure segments in the neonatal electroencephalogram (EEG) by characterizing the spectral features of the EEG waveform using a rule-based algorithm cascaded with a neural network. A rule-based algorithm screens out short segments of pseudosinusoidal EEG patterns as epileptic based on features in the power spectrum. The output of the rule-based algorithm is used to train and compare the performance of conventional feedforward neural networks and quantum neural networks. The results indicate that the trained neural networks, cascaded with the rule-based algorithm, improved the performance of the rule-based algorithm acting by itself. The evaluation of the proposed cascaded scheme for the detection of pseudosinusoidal seizure segments reveals its potential as a building block of the automated seizure detection system under development.

FEEDFORWARD NEURAL NETWORK MODELS FOR HANDLING CLASS OVERLAP AND CLASS IMBALANCE

This paper proposes a framework for training feedforward neural network models capable of handling class overlap and imbalance by minimizing an error function that compensates for such imperfections of the training set. A special case of the proposed error function can be used for training variance-controlled neural networks (VCNNs), which are developed to handle class overlap by minimizing an error function involving the class-specific variance (CSV) computed at their outputs. Another special case of the proposed error function can be used for training class-balancing neural networks (CBNNs), which are developed to handle class imbalance by relying on class-specific correction (CSC). VCNNs and CBNNs are compared with conventional feedforward neural networks (FFNNs), quantum neural networks (QNNs), and resampling techniques. The properties of VCNNs and CBNNs are illustrated by experiments on artificial data. Various experiments involving real-world data reveal the advantages offered by VCNNs and CBNNs in the presence of class overlap and class imbalance.

An evaluation of quantum neural networks in the detection of epileptic seizures in the neonatal electroencephalogram

This paper presents the results of an experimental study that evaluated the ability of quantum neural networks (QNNs) to capture and quantify uncertainty in data and compared their performance with that of conventional feedforward neural networks (FFNNs). In this work, QNNs and FFNNs were trained to classify short segments of epileptic seizures in neonatal EEG. The experiments revealed significant differences between the internal representations created by trained QNNs and FFNNs from sample information provided by the training data. The results of this experimental study also confirmed that the responses of trained QNNs are more reliable indicators of uncertainty in the input data compared with the responses of trained FFNNs.

Characterization of basic properties for pure substances and petroleum fractions by neural network

A set of conventional feedforward multilayer neural networks have been proposed to predict basic properties (e.g., critical temperature ( T c), critical pressure ( P c), critical volume ( V c), acentric factor ( ω) and molecular weight (MW)) of pure compounds and petroleum fractions based on their normal boiling point ( T b) and liquid density at 293 K. The accuracy of the method is evaluated by its application for basic property estimation of various components not used in the development of the method. Furthermore, the performance of the method is compared against the performance of the other alternatives reported as the most accurate and general methods for basic property prediction. Results of this comparison show that the proposed method outperforms the other alternatives using the same parameters, both in accuracy and generality. Furthermore, the accuracy of the proposed method is almost similar to the accuracy of the only general method despite the fact that the number of parameters required for the proposed method is less than the other alternative.