Feed-forward neural networks

One critical aspect neural network designers face today is choosing an appropriate network size for a given application. Network size involves in the case of layered neural network architectures, the number of layers in a network, the number of nodes per layer, and the number of connections. Roughly speaking, a neural network implements a nonlinear mapping of u=G(x). The mapping function G is established during a training phase where the network learns to correctly associate input patterns x to output patterns u. Given a set of training examples (x, u), there is probably an infinite number of different size networks that can learn to map input patterns x into output patterns u. The question is, which network size is more appropriate for a given problem? Unfortunately, the answer to this question is not always obvious. Many researchers agree that the quality of a solution found by a neural network depends strongly on the network size used. In general, network size affects network complexity, and learning time. It also affects the generalization capabilities of the network; that is, its ability-to produce accurate results on patterns outside its training set     

Artificial neural networks

A general neural network model is introduced. The authors begin with a discussion of models for both individual neurons and for networks of neurons. A common learning rule, i.e. backward error propagation, also known as backpropagation, is described briefly and applied to an example problem     

Artificial neural networks

Artificial neural networks are explained, and the different types are described. Three different tasks for which they are suitable are discussed. They are pattern classification and associative memory, self-organization and feature extraction, and optimization     

Twisted quaternary neural networks

The quaternary neural network (QNN) proposed by Nitta is a high‐dimensional neural network. Nitta showed that its learning is faster than that of ordinary neural networks and the number of required parameters is almost one‐third of that of real‐valued neural networks by computer simulations. In this paper, we propose the twisted quaternary neural network (TQNN) which modifies the directions of multiplications of the QNN. Since quaternions are noncommutative on multiplication, we can get another neural network. When the activation function is linear, multilayered neural networks can be expressed by single‐layered neural networks. But the TQNN cannot be expressed by a single‐layered QNN even if the activation function is linear. Therefore, the TQNN is expected to produce a variety of signal‐processing systems. We performed computer simulations to compare the QNN and the TQNN. Then we found that the latter's learning is a little faster. Moreover, computer simulation showed that the QNN tended to be trapped in local minima or plateaus but the TQNN did not. It is said that reducibility causes local minima and plateaus. We discuss the differences of reducibility between the QNN and the TQNN as well. © 2012 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

Computer-aided learning in artificial neural networks

This paper describes the development and evaluation of a computer-aided learning (CAL) package for a graduate course in artificial neural networks (ANNs). The package has been evaluated over a period of two academic years, both as an educational supplement to a conventional lecture course and as a completely self-sufficient, remotely taught course. The course is accessed via the World Wide Web (WWW). The course features Java applets for animation/demonstration purposes, employing the MATLAB computational engine for interactive examples and assignments. In an effort to provide a classroom-like environment, an interactive discussion forum is provided, along with weekly lecture summaries from the conventional lecture course. Automatically marked question pools are available for self-assessment.     

Testability in analogue cellular neural networks

This paper addresses the problem of testing an ACNN by postulating the need for including some extra hardware to render feasible a postfabrication test. the work presented here deals with a test methodology based on adding some extra circuitry to every cell of a regular ACNN. This methodology is just an initial proposal for taking advantage of the network regularity to perform a global test which can be externally interpreted and hence has potential application for reconfiguring the network.     

Artificial neural networks in power system restoration

Power system restoration (PSR) has been a subject of study for many years. Many techniques were proposed to solve the limitations of the predetermined restoration guidelines and procedures used by a majority of system operators to restore a system following the occurrence of a wide area disturbance. This paper discusses limitations encountered in some currently used PSR techniques and a proposed improvement based on artificial neural networks (ANNs). The proposed scheme is tested on a 162-bus transmission system and compared with a breadth-search restoration scheme. The results indicate that the use of ANN in power system restoration is a feasible option that should be considered for real-time applications.     

Self-organization of neural networks for clustering

Generally, there are many methods of categorizing unknown data in statistics. In many of these methods, we need sample data to determine the borders of the groups to which these data belong. Neural networks are also usable to classify unknown data. In the learning process of neural networks, we must prepare so-called teaching signals, that is, sample data. In this paper, we propose an empirical scheme to organize neural networks for clustering unknown data which belong to two groups. In our scheme, a neural network that satisfies an evaluation function without teaching signals is organized. This evaluation function is determined by a histogram of outputs of the neural network. Generally, neural networks map the input data distribution to the output distribution. Maximizing the evaluation function means separating these two output distributions from each other. As an organizing mechanism, the genetic algorithm is used because of its ability to converge to a global maximum. Some numerical results are presented to confirm the feasibility of the scheme. © 1997 Scripta Technica, Inc. Electr Eng Jpn, 121(1): 51–59, 1997     

Application of artificial neural networks in network calculations

Contents We describe radial basis functions as a special form of artificial neural networks and test whether they are applicable to power systems. In the application to distance protection, the dividing and filtering capability are in the centre of interest. In modelling of equivalent circuits for external networks, primarily the static case is important. In the dynamic case the research is started by using the E′ model and the Park equations for generator modelling. Received: 30 April 2001/Accepted: 22 June 2001     

Application of artificial neural networks in voltage stabilityassessment

Voltage stability problems have been one of the major concerns for electric utilities as a result of system heavy loading. This paper reports on an investigation on the application of ANNs in voltage stability assessment. A multilayer feedforward artificial neural network (ANN) with error backpropagation learning is proposed for calculation of voltage stability margins (VSM). Based on the energy method, a direct mapping relation between power system loading conditions and the VSMs is set up via the ANN. A systematic method for selecting the ANN's input variables was developed using sensitivity analysis. The effects of ANN's training pattern sensitivity problems were also studied by dividing system operating conditions into several loading levels based on sensitivity analysis. Extensive testing of the proposed ANN-based approach indicate its viability for power system voltage stability assessment. Simulation results on five test systems are reported in the paper     

RBF神经网络在谐波检测中的应用

有源电力滤波器补偿性能与所采用的谐波检测方式有很大的依赖关系,现有的检测方法存在精度不高、对电网频率变化比较敏感、自适应能力不强的缺点.本文提出基于RBF神经网络的谐波检测方法,具有较高的运算速度、较高的检测谐波精度,以及较强的自适应能力.     

人工神经网络在传动领域中的应用

从人工神经网络的统一描述及优越特性出发，论述其适用于传动领域的主要特征，概述了其在传动领域中的应用现状，并综述了作者的相关工作。     

Fuzzy neural network-based texture analysis of ultrasonic images

The efficacy of a novel fuzzy neural network classifier for the characterization of ultrasonic liver images based on texture analysis techniques is investigated. Classification features are extracted with the use of image texture analysis techniques such as fractal dimension texture analysis, spatial gray-level dependence matrices, gray-level difference statistics, gray-level run-length statistics, and first-order gray-level parameters. These features are fed to a neural network classifier based on geometrical fuzzy sets. Starting from the construction of the Voronoi diagram of the training patterns, an aggregation of Voronoi regions is performed, leading to the identification of larger regions belonging exclusively to one of the pattern classes. The resulting scheme is a constructive algorithm that defines fuzzy clusters of patterns. Based on observations concerning the grade of membership of the training patterns to the created regions, decision probabilities are computed through which the final classification is performed     

用于APF的神经网络自适应谐波电流检测方法

介绍了一种应用于有源电力滤波器APF(Active Power Filter)的神经网络自适应谐波电流检测方法。该方法应用自适应噪声抵消技术ANCT(Adaptive Noise Canceling Technology),将基波电流作为噪声信号,从负载电流中滤除,得到谐波电流。采用两层人工神经网络实现噪声抵消。阐述了该神经网络的构造和权值自适应调整算法,应用Matlab对该方法进行了仿真研究。仿真结果表明该方法能够实时准确地检测谐波．而且计算量小．具有较强的自适应能力。     

神经元自适应控制在心脏起博器中的应用

阐述了神经元自适应PID控制的基本原理及其在心脏起博器中的具体应用。     

Chronic stress evaluation using neural networks

Discusses noninvasively obtaining an objective index of stress using a five-layer neural network. The authors present a method to evaluate chronic stress of rats that is based on the electrocardiogram (ECG) using neural networks. As a first approach, the authors focus on whether an objective index derived from the ECG reflects the internal changes described in this paper. The results show that the index derived from the ECG reflects the internal changes, which demonstrates the usefulness of stress assessment by the proposed method. Although the present study deals with a limited experimental condition, this methodology can also be applied to other types of stress     

Short-term load forecasting using neural networks

An artificial neural network (ANN) model for short-term load forecasting (STLF) is presented. The proposed model is capable of forecasting the next 24-hour load profile at one time, as opposed to the usual ‘next one hour’ ANN models. The inputs to the ANN are load profiles of the two previous days and daily maximum and minimum temperature forecasts. The network is trained to learn the next day's load profile. Testing of the model with one year of data from the Greek interconnected power system resulted in a 2.66% average absolute forecast error.     

EEG topography recognition by neural networks

Electroencephalography (EEG) pattern-recognition studies were carried out using EEG topography (readiness potential, or RP, spatiotemporal patterns) generated the moment before voluntary movements of muscles. RPs generated prior to pronouncing syllables and controlling a joystick were studied by experiments and simulation. The spatiotemporal patterns of RPs were measured by multichannel surface electrodes pasted on the subject's scalp. Backpropagation neural networks were used for RP pattern recognition. The results show that RPs generated prior to syllable pronouncement contain some information about those syllables, and that RPs generated prior to joy stick movements contain information on the direction of intended movement. They also show that neural networks can be used to recognize EEG information and so create a new type of man-machine interface for data input.     

Myoelectric signal analysis using neural networks

It is shown that the capacity of a discrete Hopfield network for functional minimization allows it to extract the time-series parameters from a myoelectric signal (MES) at a faster rate than the previously used SLS algorithm. With a two-dimensional signal space consisting of one of the parameters and the signal power, a two-layer perceptron trained using back-propagation has been used to classify MES signals from different types of muscular contractions. The results suggest that neural networks may be suitable for MES analysis tasks and that further research in this direction is warranted.