Back Propagation Neural Network Architecture Research Articles

Expert System Approach for Soil Structure Interaction and Land Use

The application of expert systems has emerged as a cost efficient tool in civil engineering over the past decades. Use of these tools for decision making in land use will be outlined in this paper. In this study, soil-structure interaction (SSI) is discussed by using expert systems, namely, neural network (NN) approaches. This method provides a new point of view for evaluations of SSI and land use. Data from 58 local sites in California, namely, earthquake, structure, and soil property data, are used. In the expert system approach, two NN architectures are used: the back propagation NN architecture and the general regression NN (GRNN) architecture. There are 21 parameters considered as input and 4 output parameters. The four output parameters chosen are soil-to-structure rigidity ratio, period lengthening, foundation damping, and whether SSI effects can be neglected or not. The results show that the GRNN approaches are more useful and powerful for evaluation of SSI and land use.

Designing Intrusion Detection System for Web Documents Using Neural Network

Cryptographic systems are the most widely used techniques for information security. These systems however have their own pitfalls as they rely on prevention as their sole means of defense. That is why most of the organizations are attracted to the intrusion detection systems. The intrusion detection systems can be broadly categorized into two types, Anomaly and Misuse Detection systems. An anomaly-based system detects com-puter intrusions and misuse by monitoring system activity and classifying it as either normal or anomalous. Misuse detection systems can detect almost all known attack patterns; they however are hardly of any use to de-tect yet unknown attacks. In this paper, we use Neural Networks for detecting intrusive web documents avail-able on Internet. For this purpose Back Propagation Neural (BPN) Network architecture is applied that is one of the most popular network architectures for supervised learning. Analysis is carried out on Internet Security and Acceleration (ISA) server 2000 log for finding out the web documents that should not be accessed by the unau-thorized persons in an organization. There are lots of web documents available online on Internet that may be harmful for an organization. Most of these documents are blocked for use, but still users of the organization try to access these documents and may cause problem in the organization network.

Bottle-neck type of neural network as a mapping device towards food specifications

A novel methodology is proposed for food specifications associated with the origin of food. The methodology was tested on honey samples collected within the TRACE EU project. The data were sampled in various regions in Europe and analysed for the trace elements content. The sampling sites were characterized by different geological origins, such as limestone, shale, or magmatic. We have chosen 14 elements, B, Na, Mg, A, K, Ca, Mn, Co, Ni, Cu, Zn, Rb, Sr, and Ba, due to their influence on the separation of samples regarding the geology of the sampling sites. A special architecture of an error back-propagation neural network, so called bottle-neck type of neural network was used to project the data into a 2D plane. The data were fed into the 14-nodes input layer and then transferred through the 2-nodes hidden layer (compared to a bottle-neck) to the 14-nodes output layer. The two hidden nodes representing the two coordinates of the projection plane enable us to map the samples used for training of the bottle-neck network. With the knowledge about the classes of individual samples we determine the clusters in the projection plane and consequently obtain the coordinates of the centroid (gravity point) of a particular cluster. The clusters are characterized with an ellipse shape borders spanning the length of up to 3 σ in each dimension. Since the data were classified as regard to the geology, three main clusters were sought: (i) limestone, (ii) shale/mudstone/clay/loess, and (iii) acid-magmatic origin of honey samples. The novel methodology proposed for food specifications was demonstrated on a reduced set of samples, which shows good clustering of all three classes in the projection plane, and on the third class of the original data set.

Phyllometric parameters and artificial neural networks for the identification of Banksia accessions

Taxonomic identification is traditionally carried out with dichotomous keys, or at least computer-based identification keys, often on the basis of subjective visual assessment and frequently unable to detect small differences at subspecies and varietal ranks. The aims of the present work were to (1) clearly discriminate a wide group of accessions (species, subspecies and varieties) belonging to the genus Banksia on the basis of 14 phyllometric parameters determined by image analysis of the leaves, and (2) unequivocally identify the accessions with a relatively simple back-propagation neural-network (BPNN) architecture (single hidden layer) in order to develop a complementary method for fast botanical identification. The results indicate that this kind of network could be effectively and successfully used to discriminate among Banksia accessions, as the BPNN enabled a 93% unequivocal and correct simultaneous identification. Our BPNN had the advantage of being able to resolve subtle associations between characters, and of making incomplete data (i.e. absence of Banksia flower parameters such as the colour or size of styles) useful in species diagnostics. This method is relatively useful; it is easy to execute as no particular competences are necessary, equipment is low cost (scanner connected to a PC and software available as freeware) and data acquisition is fast and effective.

Feasibility of near-infrared hyperspectral imaging to differentiate Canadian wheat classes

Differentiation of wheat classes is one of the important challenges to the Canadian grain industry. Even though some wheat classes may look similar, their chemical composition and consequently the end-product quality can vary significantly. Visual differentiation of wheat classes suffers from disadvantages such as inconsistency, low throughput, and labour intensiveness. A near-infrared (NIR) hyperspectral imaging system was used to develop classification models to differentiate wheat classes grown in western Canada. Wheat bulk samples were scanned in the wavelength region of 960–1700 nm at 10 nm intervals using an InGaAs NIR camera. Seventy-five relative reflectance intensities were extracted from the scanned images and used for the differentiation of wheat classes using a statistical classifier and an artificial neural network (ANN) classifier. Classification accuracies were 100% in classifying Canada Prairie Spring Red (CPSR), Canada Western Red Winter (CWRW), and Canada Western Soft White Spring (CWSWS) wheat classes and >94% for the other wheat classes (Canada Western Extra Strong (CWES), Canada Western Hard White Spring (CWHWS), Canada Western Red Spring (CWRS), Canada Prairie Spring White (CPSW) and Canada Western Amber Durum (CWAD)) using Linear Discriminant Analysis (LDA) with a leave-one-out cross-validation method. In Quadratic Discriminant Analysis (QDA) with a leave-one-out cross-validation method, the classification accuracies were >86% for all wheat classes. The overall classification accuracies of 60% training–30% testing–10% validation (referred to as 60-30-10) and 70% training–20% testing–10% validation (referred to as 70-20-10) ANN models were above 90% for independent validation sets using three-layer standard and Wardnet back-propagation neural network architectures.

Derivative and GA-based methods in metamodeling of back-propagation neural networks for constrained approximate optimization

Artificial neural networks (ANN) have been extensively used as global approximation tools in the context of approximate optimization. ANN traditionally minimizes the absolute difference between target outputs and approximate outputs thereby resulting in approximate optimal solutions being sometimes actually infeasible when it is used as a metamodel for inequality constraint functions. The paper explores the development of the efficient back-propagation neural network (BPN)-based metamodel that ensures the constraint feasibility of approximate optimal solution. The BPN architecture is optimized via two approaches of both derivative-based method and genetic algorithm (GA) to determine interconnection weights between layers in the network. The verification of the proposed approach is examined by adopting a standard ten-bar truss problem. Finally, a GA-based approximate optimization of suspension with an optical flying head is conducted to enhance the shock resistance capability in addition to dynamic characteristics.

GA based meta-modeling of BPN architecture for constrained approximate optimization

Artificial neural networks (ANN) have been extensively used as global approximation tools in the context of approximate optimization. ANN traditionally minimizes the absolute difference between target outputs and approximate outputs, thereby resulting in approximate optimal solutions being sometimes actually infeasible when it is used as a meta-model for inequality constraint functions. The paper explores the development of the modified back-propagation neural network (BPN) based meta-model that ensures the constraint feasibility of approximate optimal solution. The BPN architecture is optimized via genetic algorithm (GA) to determine integer/continuous decision parameters such as the number of hidden layers, the number of neurons in a hidden layer, and interconnection weights between layers in the network. The verification of the proposed approach is examined by adopting a number of standard structural problems and an optical disk drive (ODD) suspension problem. Finally, GA based approximate optimization of suspension with optical flying head (OFH) is conducted to enhance the shock resistance capability in addition to dynamic characteristics.

Development of an artificial neural network-based software for prediction of power plant canal water discharge temperature

Power plant cooling water systems that interact with nearby effluents are complex non-linear, large-time-delay systems. A neural network-based software tool was developed for prediction of the canal water discharge temperature at a coal-fired power plant as a function of plant operating parameters and local weather conditions, including tide information. The plant has four units totaling an installed capacity of 1550MW and its water thermal discharge is environmentally regulated. In the summer months, when the price of electricity is very profitable and the risk of exceeding the canal temperature limit is greater, the tradeoff between maximum generation and environmental compliance violations is financially significant. The software is a predictive tool to assist in scheduling load generation among the plant's four units without exceeding a thermal discharge limit of 95^oF. Back propagation neural network architectures were trained using plant operating data with an 'off-set' component. The artificial intelligence models produced reasonable trends for year-round prediction and different operational scenarios. Comparison of measured and predicted canal temperatures indicated an accuracy of less than 0.3^oF over the range between 90 and 95^oF. The software tool was developed as an Object Linking and Embedding (OLE) for Process Control (OPC) client, with real-time communication and interface with the plant Distributed Control System (DCS).

Neural Networks as a Tool for Quality Assessment of Soybean and Rapeseed Oils Analyzed by Thermogravimetry

The quality of commercial vegetable oils is usually evaluated via chemical parameters such as density, refractive index, saponification, iodine and acid values. In this paper, the applicability of thermal parameters for the quality assessment of vegetable oils is proposed. In order to achieve this goal, different back-propagation neural network architectures were trained, using chemical and thermal parameters as inputs. To avoid any accidental correlation due to the random initialization of the weights, each topology was repeated three times and three networks were chosen, with 5-3-2, 8-5-2 and 13-6-2 structures. The error function sum square error (SSE) was used as the criterion for finalization of the learning process. A model was developed for the correct classification of oils with regard to their type and quality.

Neural network classification and novelty detection

Novel data are data belonging to classes not included in the training set of the classifier. Neural network classifiers tend to put these data into the category of patterns that most resemble the novel ones, rather than label them as novel or unknown. This work investigates the ability of the back-propagation neural network architecture to detect novel patterns and concludes that this method is unsuitable for this task. It also explores the applicability of a different neural network architecture, the probabilistic neural network, and finds that this method shows superior performance as an overall classifier when compared to back-propagation and, in addition, is able to identify novel patterns.

Neural network based temperature-dependent quantitative structure property relations (QSPRs) for predicting vapor pressure of hydrocarbons.

A neural network based quantitative structure-property relationship (QSPR) was developed for the vapor pressure-temperature behavior of hydrocarbons based on a data set for 274 compounds. The optimal QSPR model was developed based on a 7-29-1 back-propagation neural network architecture using valance molecular connectivity indices (1chi(v), 3chi(v), 4chi(v)), molecular weight, and temperature as input parameters. The average absolute errors in vapor pressure predictions for the test, validation, and overall data sets were 8.2% (0.036 log P units or 23.2 kPA), 9.2% (0.039 log P units or 26.8 kPA), and 10.7% (0.046 log P units or 31.1 kPA), respectively. The performance of the QSPR for temperature-dependent vapor pressure, which was developed from a simple set of molecular descriptors, displayed accuracy of better than or well within the range of other available estimation methods.

Risk assessment in internal auditing: a neural network approach

Risk assessment is a systematic process for integrating professional judgments about relevant risk factors, their relative significance and probable adverse conditions and/or events leading to identification of auditable activities (IIA, 1995, SIAS No. 9). Internal auditors utilize risk measures to allocate critical audit resources to compliance, operational, or financial activities within the organization (Colbert, 1995). In information rich environments, risk assessment involves recognizing patterns in the data, such as complex data anomalies and discrepancies, that perhaps conceal one or more error or hazard conditions (e.g. Coakley and Brown, 1996; Bedard and Biggs, 1991; Libby, 1985). This research investigates whether neural networks can help enhance auditors’ risk assessments. Neural networks, an emerging artificial intelligence technology, are a powerful non-linear optimization and pattern recognition tool (Haykin, 1994; Bishop, 1995). Several successful, real-world business neural network application decision aids have already been built (Burger and Traver, 1996). Neural network modeling may prove invaluable in directing internal auditor attention to those aspects of financial, operating, and compliance data most informative of high-risk audit areas, thus enhancing audit efficiency and effectiveness. This paper defines risk in an internal auditing context, describes contemporary approaches to performing risk assessments, provides an overview of the backpropagation neural network architecture, outlines the methodology adopted for conducting this research project including a Delphi study and comparison with statistical approaches, and presents preliminary results, which indicate that internal auditors could benefit from using neural network technology for assessing risk. Copyright © 1999 John Wiley & Sons, Ltd.

Crack Identification by Artificial Neural Network

Current development in smart materials has improved the conventional composite materials in the ability of self-monitor during service. The massless sensor of micro-electro-mechanical system makes the work successful to embed the sensor into the composite material without changing the original system. For an ideal on-line identification, the measured values of sensors are obtained during service not from experiment. Due to this important reason, the study of the crack identification by using the static deformation as the input is gradually an interested research which will be different from the conventional non-destructive techniques such as, ultrasonics, magnetic flux leakage, X-rays, penetrant, eddy current, etc. To achieve the on-line identification, the artificial neural network is considered instead of the usual nonlinear optimization technique. An artificial neural network is a parallel, distributed information processing structure consisting of processing elements interconnected with weights. In this paper, a most popular learning scheme called the back-propagation neural network (BPN) will be applied for implementing the crack identification. A typical architecture of BPN contains three basic layers input, output and hidden layers. According to Kolmogorov's mapping theorem, any continuous function could be mapped exactly by a three-layer feedforward neural networks. The input pattern is propagated forward, and the calculated responses are obtained. The errors between the desired outputs and the calculated outputs then propagate backward through the network, providing vital information for weight adaptation. It is known that the crack effects are localized which may not be clearly reflected

Theory of the backpropagation neural network

Publisher Summary This chapter presents a survey of the elementary theory of the basic backpropagation neural network architecture, covering the areas of architectural design, performance measurement, function approximation capability, and learning. The survey includes a formulation of the backpropagation neural network architecture to make it a valid neural network and a proof that the backpropagation mean squared error function exists and is differentiable. Also included in the survey is a theorem showing that any L2 function can be implemented to any desired degree of accuracy with a three-layer backpropagation neural network. An appendix presents a speculative neurophysiological model illustrating the way in which the backpropagation neural network architecture might plausibly be implemented in the mammalian brain for corticocortical learning between nearby regions of cerebral cortex. One of the crucial decisions in the design of the backpropagation architecture is the selection of a sigmoidal activation function.