Showing papers on "Hybrid neural network published in 2006"

A hybrid adaptive time-delay neural network model for multi-step-ahead prediction of sunspot activity

Abstract: The availability of accurate empirical models for multi-step-ahead (MS) prediction is desirable in many areas. Some ANN technologies, such as multiple-neural network, time-delay neural network (TDNN), and adaptive time-delay neural network (ATNN), have proven successful in addressing various complicated problems. The purpose of this study was to investigate the applicability of neural network MS predictive models. Motivated by the above-mentioned technologies, we proposed a hybrid neural network model, which integrated characteristics decomposition units, and a dynamic spline interpolation unit into the multiple ATNNs. Inside the net, the regular and certain information were extracted to ATNN, while both time delays and weights were dynamically adapted. The yearly average of the sunspots, which has been considered by geophysicists, environment scientists, and climatologists as a complicated non-linear system, was selected to test the hybrid model. Comparative results were presented between a traditional MS predictive model based on TDNN and the proposed model. Validation studies indicated that the proposed model is quite effective in MS prediction, especially for single-factor time series.

Neural Networks for Continuous Online Learning and Control

Abstract: This paper proposes a new hybrid neural network (NN) model that employs a multistage online learning process to solve the distributed control problem with an infinite horizon. Various techniques such as reinforcement learning and evolutionary algorithm are used to design the multistage online learning process. For this paper, the infinite horizon distributed control problem is implemented in the form of real-time distributed traffic signal control for intersections in a large-scale traffic network. The hybrid neural network model is used to design each of the local traffic signal controllers at the respective intersections. As the state of the traffic network changes due to random fluctuation of traffic volumes, the NN-based local controllers will need to adapt to the changing dynamics in order to provide effective traffic signal control and to prevent the traffic network from becoming overcongested. Such a problem is especially challenging if the local controllers are used for an infinite horizon problem where online learning has to take place continuously once the controllers are implemented into the traffic network. A comprehensive simulation model of a section of the Central Business District (CBD) of Singapore has been developed using PARAMICS microscopic simulation program. As the complexity of the simulation increases, results show that the hybrid NN model provides significant improvement in traffic conditions when evaluated against an existing traffic signal control algorithm as well as a new, continuously updated simultaneous perturbation stochastic approximation-based neural network (SPSA-NN). Using the hybrid NN model, the total mean delay of each vehicle has been reduced by 78% and the total mean stoppage time of each vehicle has been reduced by 84% compared to the existing traffic signal control algorithm. This shows the efficacy of the hybrid NN model in solving large-scale traffic signal control problem in a distributed manner. Also, it indicates the possibility of using the hybrid NN model for other applications that are similar in nature as the infinite horizon distributed control problem

Hybrid Neural Network Models Applied to a Free Radical Polymerization Process

Abstract: This article presents different ways of obtaining hybrid models, which are composed of a simplified phenomenological model and one or several neural networks. As an example, we consider free radical polymerization of methyl methacrylate, achieved through a batch bulk process, in which modeling of conversion and polymerization degrees is analyzed. Kinetics of the process is described through a simplified phenomenological model that does not take into account the gel and glass effects. This last part of the process, which is more difficult to model, is rendered by means of feed-forward neural networks with one or two hidden layers. In the present paper, the hybridization procedure is made in three ways: 1) the neural network corrects the outputs of the simplified kinetic model by modeling the residuals of conversion and polymerization degrees; 2) the neural network provides accurate values of the rate constants to the simplified kinetic model; 3) the neural network models that part of the process in which gel and glass effects appear. It is demonstrated that accurate results are obtained in all three cases, and the hybrid models are easily created and manipulated, especially because they are based on neural networks with quite simple topologies.

Intrusion detection using PCASOM neural networks

Abstract: This paper proposes a method to detect network intrusions by using the PCASOM (principal components analysis and self-organizing map) neural networks. A modified unsupervised learning algorithm which is more suitable for intrusion detection is presented. Experiments are carried out to illustrate the performance of the proposed method by using DARPA 1998 evaluation data sets. It shows that the proposed method can cluster the network connections into proper clusters with high detection rate and relatively low false alarm rate.

A Hybrid Neural Network-Data Base Correlation Positioning in GSM Network

Abstract: Mobile terminal (MT) localization in a GSM environment has been of big interest in the recent years. This work exploits the advantage of position estimations from different sources in a robust fusion algorithm to reduce the positioning error. A hybrid neural network (NN)-data base correlation method (DC) is discussed. Before the fusion process, the DC position estimates are post-processed using an extra NN in order to reduce its error. Function approximation and classification properties of the NN will be investigated and the best NN architecture will be applied in the positioning algorithm. Results show that, the post processing of the DC results has a big impact on the positioning accuracy and the fusion process gets the MT estimate within a better accuracy.

Optimal Selection of Machining Conditions in the Electrojet Drilling Process Using Hybrid NN-DF-GA Approach

Abstract: This article presents a hybrid neural network, desirability function, and genetic algorithm (NN-DF-GA) approach for optimal selection of the input process parameters for optimizing the multiresponse parameters of the electrojet drilling (EJD) process. EJD is a promising nontraditional machining technique that is used for machining microholes (<1 mm in diameter) in difficult-to-machine materials. The proposed approach first uses a back propagation neural network to formulate a fitness function for predicting the response parameters of the process. From the network output, the desirability method obtains a composite fitness function for further use in the genetic algorithm. The genetic algorithm predicts the optimal input parametric combinations and simultaneously optimizes the multiresponse characteristics of the process. Simulated results confirm the feasibility of this approach and show a good agreement with experimental results for a wide range of machining conditions.

Hybrid Neural Network

Abstract: A hybrid neural network is proposed and implemented. The proposed network is a linear hierarchical network which consists of two subnetworks. The first subnetwork is based on Kohonen Self-Organizing Map. The second subnetwork is based on Learning Vector Quantization. The hybrid neural network is tested and used for Arabic phoneme recognizer.

Retrospective optimization of time-dependent fermentation control strategies using time-independent historical data.

Abstract: We have previously shown the usefulness of historical data for fermentation process optimization. The methodology developed includes identification of important process inputs, training of an artificial neural network (ANN) process model, and ultimately use of the ANN model with a genetic algorithm to find the optimal values of each critical process input. However, this approach ignores the time-dependent nature of the system, and therefore, does not fully utilize the available information within a database. In this work, we propose a method for incorporating time-dependent optimization into our previously developed three-step optimization routine. This is achieved by an additional step that uses a fermentation model (consisting of coupled ordinary differential equations (ODE)) to interpret important time-course features of the collected data through adjustments in model parameters. Important process variables not explicitly included in the model were then identified for each model parameter using automatic relevance determination (ARD) with Gaussian process (GP) models. The developed GP models were then combined with the fermentation model to form a hybrid neural network model that predicted the time-course activity of the cell and protein concentrations of novel fermentation conditions. A hybrid-genetic algorithm was then used in conjunction with the hybrid model to suggest optimal time-dependent control strategies. The presented method was implemented upon an E. coli fermentation database generated in our laboratory. Optimization of two different criteria (final protein yield and a simplified economic criteria) was attempted. While the overall protein yield was not increased using this methodology, we were successful in increasing a simplified economic criterion by 15% compared to what had been previously observed. These process conditions included using 35% less arabinose (the inducer) and 33% less typtone in the media and reducing the time required to reach the maximum protein concentration by 10% while producing approximately the same level of protein as the previous optimum.

A modified hybrid neural network for pattern recognition and its application to SSW complex in EEG

Abstract: In this study, a modified hybrid neural network with asymmetric basis functions is presented for feature extraction of spike and slow wave complexes in electroencephalography (EEG). Feature extraction process has a great importance in all pattern recognition and classification problems. A gradient descent algorithm, indeed a back propagation type, is adapted to the proposed artificial neural network. The performance of the proposed network is measured using a support vector machine classifier fed by features extracted using the proposed neural network. The results show that the proposed neural network model can effectively be used in pattern recognition tasks. In experiments, real EEG data are used.

Short Term Load Forecasting Using a Hybrid Neural Network

Abstract: Short term load forecasting (STLF) is very important from the power systems grid operation point of view. STLF involves forecasting load demand in a short term time frame. The short term time frame may consist of half hourly prediction up to weekly prediction. Accurate forecasting would benefit the utility in terms of reliability and stability of the grid ensuring adequate supply is present to meet with the load demand. Apart from that it would also affect the financial performance of the utility company. An accurate forecast would result in better savings while maintaining the security of the grid. This paper outlines the STLF using a novel hybrid online learning neural network, known as the Gaussian regression (GR). This new hybrid neural network is a combination of two existing online learning neural networks which are the Gaussian adaptive resonance theory (GA) and the generalized regression neural network (GRNN). Both GA and GRNN implemented online learning, but each of them suffers from limitation. Originally GA is used for unsupervised clustering by compressing the training samples into several categories. A supervised version of GA is available, namely Gaussian ARTMAP (GAM). However, the GAM is still not capable on solving regression problem. On the other hand, GRNN is designed for solving real value estimation (regression) problem, but the learning process would involve of memorizing all training samples, hence high computational cost. The hybrid GR is considered an enhanced version of GRNN with compression ability while still maintains online learning properties. Simulation results show that GR has comparable prediction accuracy and has less prototype as compared to the original GRNN as well as the support vector regression.

A hybrid unscented kalman filter and support vector machine model in option price forecasting

Abstract: This study develops a hybrid model that combines unscented Kalman filters (UKFs) and support vector machines (SVMs) to implement an online option price predictor. In the hybrid model, the UKF is used to infer latent variables and make a prediction based on the Black-Scholes formula, while the SVM is employed to capture the nonlinear residuals between the actual option prices and the UKF predictions. Taking option data traded in Taiwan Futures Exchange, this study examined the forecasting accuracy of the proposed model, and found that the new hybrid model is superior to pure SVM models or hybrid neural network models in terms of three types of options. This model can also help investors for reducing their risk in online trading.

Hybrid neural network model based on multi-layer perceptron and adaptive resonance theory

Abstract: The model of the hybrid neural network is considered. This model consists of model ART-2 for clustering and perceptron for preprocessing of images. The perceptron provides invariant recognition of objects. This model can be used in mobile robots for recognition of new objects or scenes in sight the robot during his movement.

Hybrid Neural Network Model Based on Multi-layer Perceptron and Adaptive Resonance Theory

Abstract: The model of the hybrid neural network is considered. This model consists of model ART-2 for clustering and perceptron for preprocessing of images. The perceptron provides invariant recognition of objects. This model can be used in mobile robots for recognition of new objects or scenes in sight the robot during his movement.

Optimized column design using genetic algorithm based neural networks

Abstract: In the structural design of columns, the dimensions of the column and reinforcement are initially assumed and then the interaction formula is used to verify the suitability of chosen dimensions and reinforcement. This approach necessitates few trials for coming up with an economical and safe design. This paper demonstrates the applicability of artificial neural networks (ANN) and genetic algorithms (GA) for the design of short columns under biaxial bending. A hybrid neural network model that combines the features of feed forward neural networks and genetic algorithms has been developed for the design of short column subjected to biaxial bending. The network has been trained with design data obtained from design experts in the field. The hybrid neural network model learned the design of column in just 1800 training cycles. After successful learning, the model predicted the percentage of steel required for new problems with good accuracy satisfying all design constraints. The results of the hybrid network model are compared with the solution of an optimizer, which uses the interior penalty function method. The various stages involved in the development of genetic algorithm based neural network model are addressed.

Hybrid neural network design and implementation on FPGA for infant cry recognition

Abstract: It has been found that the infant's crying has much information on its sound wave. For small infants crying is a form of communication, a very limited one, but similar to the way adults communicate. In this work we present the design of an Automatic Infant Cry Recognizer hybrid system, that classifies different kinds of cries, with the objective of identifying some pathologies in recently born babies. The system is based on the implementation of a Fuzzy Relational Neural Network (FRNN) model on a standard reconfigurable hardware like Field Programmable Gate Arrays (FPGAs). To perform the experiments, a set of crying samples is divided in two parts; the first one is used for training and the other one for testing. The input features are represented by fuzzy membership functions and the links between nodes, instead of regular weights, are represented by fuzzy relations. The training adjusts the relational weight matrix, and once its values have been adapted, the matrix is fixed into the FPGA. The goal of this research is to prove the performance of the FRNN in a development board; in this case we used the RC100 from Celoxica. The implementation process, as well as some results is shown.

An Intrusion Detection System for Network-Initiated Attacks Using a Hybrid Neural Network

Abstract: We present a hybrid system based on a combination of Neural Networks and rule-based matching systems that is capable of detecting network-initiated intrusion attacks on web servers. The system has a strong learning component allowing it to recognize even novel attacks (i.e. attacks it has never seen before) and categorize them as such. The performance of the Neural Network in detecting attacks is very good with success rates of more than 78% in recognizing new attacks. However, because of an alarmingly high false alarm rate that measures more than 90% on normal HTTP traffic carrying image uploads we had to combine the original ANN with a rule-based component that monitors the server’s system calls for detecting unusual activity. A final component combines the two systems to make the final decision on whether to raise an intrusion alarm or not. We report on the results we got from our approach and future directions for this research.

Hybrid neural network design and implementation on FPGA for infant cry recognition

Abstract: It has been found that the infant's crying has much information on its sound wave For small infants crying is a form of communication, a very limited one, but similar to the way adults communicate In this work we present the design of an Automatic Infant Cry Recognizer hybrid system, that classifies different kinds of cries, with the objective of identifying some pathologies in recently born babies The system is based on the implementation of a Fuzzy Relational Neural Network (FRNN) model on a standard reconfigurable hardware like Field Programmable Gate Arrays (FPGAs) To perform the experiments, a set of crying samples is divided in two parts; the first one is used for training and the other one for testing The input features are represented by fuzzy membership functions and the links between nodes, instead of regular weights, are represented by fuzzy relations The training adjusts the relational weight matrix, and once its values have been adapted, the matrix is fixed into the FPGA The goal of this research is to prove the performance of the FRNN in a development board; in this case we used the RC100 from Celoxica The implementation process, as well as some results is shown.

Hybrid Neural Network Bankruptcy Prediction: An Integration of Financial Ratios, Intellectual Capital Ratios, MDA, and Neural Network Learning

Abstract: One purpose of this paper is to propose the hybrid neural network models for bankruptcy prediction The proposed hybrid neural network models are, respectively, a MDA model integrated with financial ratios, a MDA model integrated with financial ratios and intellectual capital ratios, a MDA-assisted neural network model integrated with financial ratios, and a MDA-assisted neural network model integrated with financial ratios and intellectual capital ratios The performance of the hybrid neural network model is compared with MDA model integrated with financial ratios as a benchmark Empirical results using Taiwan bankruptcy data show that hybrid neural network models are very promising ones in terms of accuracy and adaptability

Viscosity Prediction for PET Process Based on Hybrid Neural Networks

Abstract: Based on the "divide and rule" idea, hybrid neural networks (HNNs), which consisted of linear dynamic neural network and nonlinear static neural network, was used to model for complicated nonlinear systems. By using hybrid neural networks, it can reduce the degree of difficulty for training a single network, e.g. long training time and lower accuracy; and also can transfer the solution of nonlinear control strategy into solving for linear systems based on the decomposed models. An industrial polymerization process was introduced as a powerful case-study for the demonstration of potential of neural modeling. Nonlinear predicative models, based on both serial and parallel neural networks, were applied to predict the dynamic viscosity of PET. And the results indicated that both parallel and serial hybrid neural networks can model for complicated systems well

Intelligent Built-in Test Fault Diagnosis Based on Wavelet Analysis and Neural Networks

Abstract: This paper proposes a new intelligent Built-in Test (BIT) fault diagnosis system based on wavelet analysis and neural networks. The aim of this investigation is to improve the fault diagnosis capability of intelligent BIT for More-Electric Aircraft Electrical Power System (MEAEPS). In constructing the BIT system, the wavelet packet transform is applied to extract fault features. Through the wavelet packet decomposition, we get the fault eigenvectors and input them into a hybrid neural network, which performs in the role of a fault classifier. This hybrid network adds a supervised Learning Vector Quantization (LVQ) layer to the Generalized Learning Vector Quantization (GLVQ) network, which makes the boundaries among the fault classes more discriminative than using the GLVQ network alone. Since the original GLVQ algorithm suffers from several major problems, we modify the original algorithm in order to make this network more suitable for application. This modified algorithm employs a new form of loss factor, and its learning rules are derived through finding a minimum of the loss function. Finally, the proposed method has been applied to the BIT system of the MEAEPS, and the results have shown that the proposed method is promising to improve the performance of the intelligent BIT system.

Face detection using an adaptive skin-color filter and FMM neural networks

Abstract: In this paper, we present a real-time face detection method based on hybrid neural networks. We propose a modified version of fuzzy min-max (FMM) neural network for feature analysis and face classification. A relevance factor between features and pattern classes is defined to analyze the saliency of features. The measure can be utilized for the feature selection to construct an adaptive skin-color filter. The feature extraction module employs a convolutional neural network (CNN) with a Gabor transform layer to extract successively larger features in a hierarchical set of layers. In this paper we first describe the behavior of the proposed FMM model, and then introduce the feature analysis technique for skin-color filter and pattern classifier.

Recognition of fatty liver using hybrid neural network

Abstract: A hybrid neural network based on self-organizing map (SOM) and multilayer perception(MLP) artificial neural network(ANN) is proposed for recognition of fatty liver from B-scan ultrasonic images. Firstly, four texture features including angular second moment, contrast, entropy and inverse differential moment were extracted from gray-level co-occurrence matrices of B-scan ultrasound liver images. They were mapped by a SOM for feature reduction, and then combined with other two features, named approximate entropy and mean intensity ratio. All features were imposed to a MLP for recognition. In the experiment, 130 B-scan liver images were divided into two groups: 104 in training group and 26 in validation group. Both the normal and fatty livers were recognized correctly. This study showed that the hybrid neural network could be used for fatty liver recognition with good performances.

Implementation of a neural network processor based on RISC architecture for various signal processing applications

Abstract: In this paper, hybrid neural network processor (HANNP) is designed in VLSI. The HANNP has RISC based architecture leading to an effective general digital signal processing and artificial neural networks computation. The architecture of a HANNP including the general digital processing units such as 64-bit floating-point arithmetic unit (FPU), a control unit (CU) and neural network processing units such as artificial neural computing unit (NNPU), specialized neural data bus and interface unit, etc. The HANNP is modeled in Veilog HDL and implemented with FPGA. Character recognition problems and Kohonen self-organization problems are applied to the proposed HANNP to justify its applicability to real engineering problems.

Intelligent Built-in Test (BIT) for More-Electric Aircraft Power System Based on Hybrid Generalized LVQ Neural Network

Abstract: This paper proposes a hybrid neural network model based on the Generalized Learning Vector Quantization(GLVQ) learning algorithm and applies this proposed method to the BIT system of More-Electric Aircraft Electrical Power System (MEAEPS). This paper first discusses the feasibility of application unsupervised neural networks to the BIT system and the representative Generalized LVQ (GLVQ) neural network is selected due to its good performance in clustering analysis. Next, we adopt a new form of loss factor to modify the original GLVQ algorithm in order to make it more suitable for our application. Since unsupervised networks cannot distinguish the similar classes, we add a LVQ layer to the GLVQ network to construct a hybrid neural network model. Finally, the proposed method has been applied to the intelligent BIT system of the MEAEPS, and the results show that the proposed method is promising to improve the performance of the BIT system.

A Hybrid Neural Network/Analog Model for Climate Downscaling

Abstract: Synoptic downscaling models are used in climatology to model local-scale climate variables from synoptic-scale atmospheric circulation variables. This paper presents a hybrid method for multi-site downscaling that combines an artificial neural network and an analog, i.e., k-nearest neighbor, model. The method can resolve complicated synoptic-to local-scale relationships while preserving spatial relationships between sites. Performance on both synthetic and real-world datasets indicates that the hybrid model is capable of outperforming other forms of analog models used in synoptic downscaling.