Abstract: The competitive learning is an adaptive process in
which the neurons in a neural network gradually become sensitive to
different input pattern clusters. The basic idea behind the Kohonen-s
Self-Organizing Feature Maps (SOFM) is competitive learning.
SOFM can generate mappings from high-dimensional signal spaces
to lower dimensional topological structures. The main features of this
kind of mappings are topology preserving, feature mappings and
probability distribution approximation of input patterns. To overcome
some limitations of SOFM, e.g., a fixed number of neural units and a
topology of fixed dimensionality, Growing Self-Organizing Neural
Network (GSONN) can be used. GSONN can change its topological
structure during learning. It grows by learning and shrinks by
forgetting. To speed up the training and convergence, a new variant
of GSONN, twin growing cell structures (TGCS) is presented here.
This paper first gives an introduction to competitive learning, SOFM
and its variants. Then, we discuss some GSONN with fixed
dimensionality, which include growing cell structures, its variants
and the author-s model: TGCS. It is ended with some testing results
comparison and conclusions.
Abstract: It is well known that a linear dynamic system including
a delay will exhibit limit cycle oscillations when a bang-bang sensor
is used in the feedback loop of a PID controller. A similar behaviour
occurs when a delayed feedback signal is used to train a neural
network. This paper develops a method of predicting this behaviour
by linearizing the system, which can be shown to behave in a manner
similar to an integral controller. Using this procedure, it is possible
to predict the characteristics of the neural network driven limit cycle
to varying degrees of accuracy, depending on the information known
about the system. An application is also presented: the intelligent
control of a spark ignition engine.
Abstract: The ElectroEncephaloGram (EEG) is useful for
clinical diagnosis and biomedical research. EEG signals often
contain strong ElectroOculoGram (EOG) artifacts produced
by eye movements and eye blinks especially in EEG recorded
from frontal channels. These artifacts obscure the underlying
brain activity, making its visual or automated inspection
difficult. The goal of ocular artifact removal is to remove
ocular artifacts from the recorded EEG, leaving the underlying
background signals due to brain activity. In recent times,
Independent Component Analysis (ICA) algorithms have
demonstrated superior potential in obtaining the least
dependent source components. In this paper, the independent
components are obtained by using the JADE algorithm (best
separating algorithm) and are classified into either artifact
component or neural component. Neural Network is used for
the classification of the obtained independent components.
Neural Network requires input features that exactly represent
the true character of the input signals so that the neural
network could classify the signals based on those key
characters that differentiate between various signals. In this
work, Auto Regressive (AR) coefficients are used as the input
features for classification. Two neural network approaches
are used to learn classification rules from EEG data. First, a
Polynomial Neural Network (PNN) trained by GMDH (Group
Method of Data Handling) algorithm is used and secondly,
feed-forward neural network classifier trained by a standard
back-propagation algorithm is used for classification and the
results show that JADE-FNN performs better than JADEPNN.
Abstract: This research work is aimed at speech recognition
using scaly neural networks. A small vocabulary of 11 words were
established first, these words are “word, file, open, print, exit, edit,
cut, copy, paste, doc1, doc2". These chosen words involved with
executing some computer functions such as opening a file, print
certain text document, cutting, copying, pasting, editing and exit.
It introduced to the computer then subjected to feature extraction
process using LPC (linear prediction coefficients). These features are
used as input to an artificial neural network in speaker dependent
mode. Half of the words are used for training the artificial neural
network and the other half are used for testing the system; those are
used for information retrieval.
The system components are consist of three parts, speech
processing and feature extraction, training and testing by using neural
networks and information retrieval.
The retrieve process proved to be 79.5-88% successful, which is
quite acceptable, considering the variation to surrounding, state of
the person, and the microphone type.
Abstract: In this paper, a neural network technique is applied to
real-time classifying media while a projectile is penetrating through
them. A laboratory-scaled penetrating setup was built for the
experiment. Features used as the network inputs were extracted from
the acceleration of penetrator. 6000 set of features from a single
penetration with known media and status were used to train the neural
network. The trained system was tested on 30 different penetration
experiments. The system produced an accuracy of 100% on the
training data set. And, their precision could be 99% for the test data
from 30 tests.
Abstract: The main objective of Automatic Generation Control (AGC) is to balance the total system generation against system load losses so that the desired frequency and power interchange with neighboring systems is maintained. Any mismatch between generation and demand causes the system frequency to deviate from its nominal value. Thus high frequency deviation may lead to system collapse. This necessitates a very fast and accurate controller to maintain the nominal system frequency. This paper deals with a novel approach of artificial intelligence (AI) technique called Hybrid Neuro-Fuzzy (HNF) approach for an (AGC). The advantage of this controller is that it can handle the non-linearities at the same time it is faster than other conventional controllers. The effectiveness of the proposed controller in increasing the damping of local and inter area modes of oscillation is demonstrated in a two area interconnected power system. The result shows that intelligent controller is having improved dynamic response and at the same time faster than conventional controller.
Abstract: In this work, a Modified Functional Link Artificial
Neural Network (M-FLANN) is proposed which is simpler than a
Multilayer Perceptron (MLP) and improves upon the universal
approximation capability of Functional Link Artificial Neural
Network (FLANN). MLP and its variants: Direct Linear Feedthrough
Artificial Neural Network (DLFANN), FLANN and
M-FLANN have been implemented to model a simulated Water Bath
System and a Continually Stirred Tank Heater (CSTH). Their
convergence speed and generalization ability have been compared.
The networks have been tested for their interpolation and
extrapolation capability using noise-free and noisy data. The results
show that M-FLANN which is computationally cheap, performs
better and has greater generalization ability than other networks
considered in the work.
Abstract: In this paper back-propagation artificial neural network
(BPANN) is employed to predict the deformation of the upsetting
process. To prepare a training set for BPANN, some finite element
simulations were carried out. The input data for the artificial neural
network are a set of parameters generated randomly (aspect ratio d/h,
material properties, temperature and coefficient of friction). The
output data are the coefficient of polynomial that fitted on barreling
curves. Neural network was trained using barreling curves generated
by finite element simulations of the upsetting and the corresponding
material parameters. This technique was tested for three different
specimens and can be successfully employed to predict the
deformation of the upsetting process
Abstract: The performance of a sucrose-based H2 production in
a completely stirred tank reactor (CSTR) was modeled by neural
network back-propagation (BP) algorithm. The H2 production was
monitored over a period of 450 days at 35±1 ºC. The proposed model
predicts H2 production rates based on hydraulic retention time
(HRT), recycle ratio, sucrose concentration and degradation, biomass
concentrations, pH, alkalinity, oxidation-reduction potential (ORP),
acids and alcohols concentrations. Artificial neural networks (ANNs)
have an ability to capture non-linear information very efficiently. In
this study, a predictive controller was proposed for management and
operation of large scale H2-fermenting systems. The relevant control
strategies can be activated by this method. BP based ANNs modeling
results was very successful and an excellent match was obtained
between the measured and the predicted rates. The efficient H2
production and system control can be provided by predictive control
method combined with the robust BP based ANN modeling tool.
Abstract: Although lots of research work has been done for
human pose recognition, the view-point of cameras is still critical
problem of overall recognition system. In this paper, view-point
insensitive human pose recognition is proposed. The aims of the
proposed system are view-point insensitivity and real-time processing.
Recognition system consists of feature extraction module, neural
network and real-time feed forward calculation. First, histogram-based
method is used to extract feature from silhouette image and it is
suitable for represent the shape of human pose. To reduce the
dimension of feature vector, Principle Component Analysis(PCA) is
used. Second, real-time processing is implemented by using Compute
Unified Device Architecture(CUDA) and this architecture improves
the speed of feed-forward calculation of neural network. We
demonstrate the effectiveness of our approach with experiments on
real environment.
Abstract: The Multi-Layered Perceptron (MLP) Neural
networks have been very successful in a number of signal processing
applications. In this work we have studied the possibilities and the
met difficulties in the application of the MLP neural networks for the
prediction of daily solar radiation data. We have used the Polack-Ribière algorithm for training the neural networks. A comparison, in
term of the statistical indicators, with a linear model most used in
literature, is also performed, and the obtained results show that the
neural networks are more efficient and gave the best results.
Abstract: In this paper back-propagation artificial neural
network (BPANN) with Levenberg–Marquardt algorithm is
employed to predict the limiting drawing ratio (LDR) of the deep
drawing process. To prepare a training set for BPANN, some finite
element simulations were carried out. die and punch radius, die arc
radius, friction coefficient, thickness, yield strength of sheet and
strain hardening exponent were used as the input data and the LDR
as the specified output used in the training of neural network. As a
result of the specified parameters, the program will be able to
estimate the LDR for any new given condition. Comparing FEM and
BPANN results, an acceptable correlation was found.
Abstract: In the present work, we propose a new technique to
enhance the learning capabilities and reduce the computation
intensity of a competitive learning multi-layered neural network
using the K-means clustering algorithm. The proposed model use
multi-layered network architecture with a back propagation learning
mechanism. The K-means algorithm is first applied to the training
dataset to reduce the amount of samples to be presented to the neural
network, by automatically selecting an optimal set of samples. The
obtained results demonstrate that the proposed technique performs
exceptionally in terms of both accuracy and computation time when
applied to the KDD99 dataset compared to a standard learning
schema that use the full dataset.
Abstract: Embedded systems need to respect stringent real
time constraints. Various hardware components included in such
systems such as cache memories exhibit variability and therefore
affect execution time. Indeed, a cache memory access from an
embedded microprocessor might result in a cache hit where the
data is available or a cache miss and the data need to be fetched
with an additional delay from an external memory. It is therefore
highly desirable to predict future memory accesses during
execution in order to appropriately prefetch data without incurring
delays. In this paper, we evaluate the potential of several artificial
neural networks for the prediction of instruction memory
addresses. Neural network have the potential to tackle the nonlinear
behavior observed in memory accesses during program
execution and their demonstrated numerous hardware
implementation emphasize this choice over traditional forecasting
techniques for their inclusion in embedded systems. However,
embedded applications execute millions of instructions and
therefore millions of addresses to be predicted. This very
challenging problem of neural network based prediction of large
time series is approached in this paper by evaluating various neural
network architectures based on the recurrent neural network
paradigm with pre-processing based on the Self Organizing Map
(SOM) classification technique.
Abstract: Recently, a lot of attention has been devoted to
advanced techniques of system modeling. PNN(polynomial neural
network) is a GMDH-type algorithm (Group Method of Data
Handling) which is one of the useful method for modeling nonlinear
systems but PNN performance depends strongly on the number of
input variables and the order of polynomial which are determined by
trial and error. In this paper, we introduce GPNN (genetic
polynomial neural network) to improve the performance of PNN.
GPNN determines the number of input variables and the order of all
neurons with GA (genetic algorithm). We use GA to search between
all possible values for the number of input variables and the order of
polynomial. GPNN performance is obtained by two nonlinear
systems. the quadratic equation and the time series Dow Jones stock
index are two case studies for obtaining the GPNN performance.
Abstract: Fast forecasting of stock market prices is very important for
strategic planning. In this paper, a new approach for fast forecasting of
stock market prices is presented. Such algorithm uses new high speed
time delay neural networks (HSTDNNs). The operation of these
networks relies on performing cross correlation in the frequency
domain between the input data and the input weights of neural
networks. It is proved mathematically and practically that the number
of computation steps required for the presented HSTDNNs is less
than that needed by traditional time delay neural networks
(TTDNNs). Simulation results using MATLAB confirm the
theoretical computations.
Abstract: A feed-forward, back-propagation Artificial Neural
Network (ANN) model has been used to forecast the occurrences of
wastewater overflows in a combined sewerage reticulation system.
This approach was tested to evaluate its applicability as a method
alternative to the common practice of developing a complete
conceptual, mathematical hydrological-hydraulic model for the
sewerage system to enable such forecasts. The ANN approach
obviates the need for a-priori understanding and representation of the
underlying hydrological hydraulic phenomena in mathematical terms
but enables learning the characteristics of a sewer overflow from the
historical data.
The performance of the standard feed-forward, back-propagation
of error algorithm was enhanced by a modified data normalizing
technique that enabled the ANN model to extrapolate into the
territory that was unseen by the training data. The algorithm and the
data normalizing method are presented along with the ANN model
output results that indicate a good accuracy in the forecasted sewer
overflow rates. However, it was revealed that the accurate
forecasting of the overflow rates are heavily dependent on the
availability of a real-time flow monitoring at the overflow structure
to provide antecedent flow rate data. The ability of the ANN to
forecast the overflow rates without the antecedent flow rates (as is
the case with traditional conceptual reticulation models) was found to
be quite poor.
Abstract: Drying characteristics of rough rice (variety of lenjan) with an initial moisture content of 25% dry basis (db) was studied in a hot air dryer assisted by infrared heating. Three arrival air temperatures (30, 40 and 500C) and four infrared radiation intensities (0, 0.2 , 0.4 and 0.6 W/cm2) and three arrival air speeds (0.1, 0.15 and 0.2 m.s-1) were studied. Bending strength of brown rice kernel, percentage of cracked kernels and time of drying were measured and evaluated. The results showed that increasing the drying arrival air temperature and radiation intensity of infrared resulted decrease in drying time. High bending strength and low percentage of cracked kernel was obtained when paddy was dried by hot air assisted infrared dryer. Between this factors and their interactive effect were a significant difference (p
Abstract: This paper uses the radial basis function neural
network (RBFNN) for system identification of nonlinear systems.
Five nonlinear systems are used to examine the activity of RBFNN in
system modeling of nonlinear systems; the five nonlinear systems are
dual tank system, single tank system, DC motor system, and two
academic models. The feed forward method is considered in this
work for modelling the non-linear dynamic models, where the KMeans
clustering algorithm used in this paper to select the centers of
radial basis function network, because it is reliable, offers fast
convergence and can handle large data sets. The least mean square
method is used to adjust the weights to the output layer, and
Euclidean distance method used to measure the width of the Gaussian
function.
Abstract: Quality control charts are very effective in detecting
out of control signals but when a control chart signals an out of
control condition of the process mean, searching for a special cause
in the vicinity of the signal time would not always lead to prompt
identification of the source(s) of the out of control condition as the
change point in the process parameter(s) is usually different from the
signal time. It is very important to manufacturer to determine at what
point and which parameters in the past caused the signal. Early
warning of process change would expedite the search for the special
causes and enhance quality at lower cost. In this paper the quality
variables under investigation are assumed to follow a multivariate
normal distribution with known means and variance-covariance
matrix and the process means after one step change remain at the new
level until the special cause is being identified and removed, also it is
supposed that only one variable could be changed at the same time.
This research applies artificial neural network (ANN) to identify the
time the change occurred and the parameter which caused the change
or shift. The performance of the approach was assessed through a
computer simulation experiment. The results show that neural
network performs effectively and equally well for the whole shift
magnitude which has been considered.