Neural network control for DC motor micromaneuvering

The application of a neural network controller for compensating the effects induced by the friction in a DC motor micromaneuvering system is considered in this article. A backpropagation neural network operating in the specialized learning mode, using the sign gradient descent algorithm, is employed. The input vector to the neural network controller consists of the time history of the motor angular shaft velocity within a prespecified time window. The on-line training of the neural network is performed in the region of interest of the output domain. The neural network output resembles that of a pulse width modulated controller. The effect of the number of neurons in the input and hidden layers on the transient system response is explored. Experimental studies are presented to indicate the effectiveness of the proposed algorithm     

Aggregate volumetric estimation based on PCA and momentum-enhanced BP neural network

This paper proposes a Back Propagation (BP) neural network with momentum enhance-ment aiming to achieving the smooth convergence for aggregate volumetric estimation purpose. Net-work inputs are first selected by optically measuring the eight geometry-related parameters from the given particle image. To simplify the network structure, principal component analysis technique is applied to reduce the input dimension. The specific network structure is finalized based on both em-pirical expertise and analysis on selecting the appropriate number of neurons in hidden layer. The network is trained using the finite number of randomly-picked particles. The training and test results suggest that, compared to the generic BP network, the training duration of the proposed neural network is greatly attenuated, the complexity of the network structure is largely reduced, and the es-timation precision is within 2%, being sufficiently up to technical satisfaction.     

Neuromorphic control: adaptation and learning

A structure for a neural network-based robotic motion controller is presented. Simulations of both position and force servos are carried out, and the approach is shown to be useful for a nonlinear system in an uncertain environment. The neural network comprises a four-layer network, including input/output layers and two hidden layers. Time delay elements are included in the first hidden layer, so that the neural network can learn dynamics of the system. The authors also implement a new learning method based on fuzzy logic, which is useful to accelerate learning and improve convergence     

An artificial neural network for SPECT image reconstruction

An artificial neural network has been developed to reconstruct quantitative single photon emission computed tomographic (SPECT) images. The network is trained with an ideal projection-image pair to learn a shift-invariant weighting (filter) for the projections. Once trained, the network produces weighted projections as a hidden layer when acquired projection data are presented to its input. This hidden layer is then backprojected to form an image as the network output. The learning algorithm adjusts the weighting coefficients using a backpropagation algorithm which minimizes the mean squared error between the ideal training image and the reconstructed training image. The response of the trained network to an impulse projection resembles the ramp filter typically used with backprojection, and reconstructed images are similar to filtered backprojection images.     

Modeling stripline discontinuities by neural network with knowledge-based neurons

A three-layer neural network with knowledge-based neurons in the hidden layer (NNKBN) is presented for modeling stripline discontinuities. In NNKBN, prior knowledge for stripline discontinuity is incorporated into each hidden neuron. With knowledge-based neurons, the learning ability and generalization of the neural network are improved. Compared with conventional multi-layer perceptron neural network, the NNKBN can map the input-output relationships with fewer hidden neurons and has higher reliability for extrapolation beyond training data range. Two examples are given to illustrate the potential power of this approach.     

Recurrent-Fuzzy-Neural-Network-Controlled Linear Induction Motor Servo Drive Using Genetic Algorithms

A recurrent fuzzy neural network (RFNN) controller based on real-time genetic algorithms (GAs) is developed for a linear induction motor (LIM) servo drive in this paper. First, the dynamic model of an indirect field-oriented LIM servo drive is derived. Then, an online training RFNN with a backpropagation algorithm is introduced as the tracking controller. Moreover, to guarantee the global convergence of tracking error, a real-time GA is developed to search the optimal learning rates of the RFNN online. The GA-based RFNN control system is proposed to control the mover of the LIM for periodic motion. The theoretical analyses for the proposed GA-based RFNN controller are described in detail. Finally, simulated and experimental results show that the proposed controller provides high-performance dynamic characteristics and is robust with regard to plant parameter variations and external load disturbance     

切比雪夫函数型连接神经网络在信道均衡中的应用

本文提出一种基于切比雪夫函数型连接神经网络(CFLNN)的信道均衡方法。传统的前馈神经网络虽然能有效地解决信道均衡的问题,但具有计算复杂度过高,收敛速度慢等缺点。函数型连接神经网络通过对输入模式进行非线性扩展,可以不必使用隐层而不降低整体性能,从而极大简化了网络结构。同时,神经网络的学习方法得以简化,提高了收敛速度。本文采用可变尺度共扼梯度下降法(SCG)对该函数型连接网络进行训练。仿真结果表明了用切比雪夫函数型连接神经网络解决信道均衡问题的有效性。     

Supervised and unsupervised learning in radial basis functionclassifiers

The paper considers a number of strategies for training radial basis function (RBF) classifiers. A benchmark problem is constructed using ten-dimensional input patterns which have to be classified into one of three classes. The RBF networks are trained using a two-phase approach (unsupervised clustering for the first layer followed by supervised learning for the second layer), error backpropagation (supervised learning for both layers) and a hybrid approach. It is shown that RBF classifiers trained with error backpropagation give results almost identical to those obtained with a multilayer perceptron. Although networks trained with the two-phase approach give slightly worse classification results, it is argued that the hidden-layer representation of such networks is much more powerful, especially if it is encoded in the form of a Gaussian mixture model. During training, the number of subclusters present within the training database can be estimated: during testing, the activities in the hidden layer of the classification network can be used to assess the novelty of input patterns and thereby help to validate network outputs     

基于GA-BP神经网络的矿用刮板输送机故障诊断

针对矿用刮板输送机的故障诊断问题,提出一种基于GA-BP神经网络的故障诊断方法。为了避免BP神经网络易陷入局部最小值、隐含层节点数难确定等问题,这里首先根据经验公式缩小隐含层节点数范围,在小范围里寻找最优的隐含层节点数;进而根据遗传算法具有全局寻优的特点,用遗传算法优化BP神经网络训练的初始权值阈值。研究表明经经验公式寻找最优隐含层节点数后,再将遗传算法与BP神经网络结合可以有效地解决神经网络收敛速度慢,易陷入局部最小等问题,提高了刮板输送机传动部的故障诊断精度。通过仿真实验验证了文中方法的有效性。     

Parallel,self-organizing,hierarchical neural networks with forward-backward training

A forward-backward training algorithm for parallel, self-organizing hierarchical neural networks (PSHNNs) is described. Using linear algebra, it is shown that the forward-backward training of ann-stage PSHNN until convergence is equivalent to the pseudo-inverse solution for a single, total network designed in the least-squares sense with the total input vector consisting of the actual input vector and its additional nonlinear transformations. These results are also valid when a single long input vector is partitioned into smaller length vectors. A number of advantages achieved are: small modules for easy and fast learning, parallel implementation of small modules during testing, faster convergence rate, better numerical error-reduction, and suitability for learning input nonlinear transformations by other neural networks. The backpropagation (BP) algorithm is proposed for learning input nonlinearitics. Better performance in terms of deeper minimum of the error function and faster convergence rate is achieved when a single BP network is replaced by a PSHNN of equal complexity in which each stage is a BP network of smaller complexity than the single BP network.     

An optimal neural network process model for plasma etching

Neural network models of semiconductor processes have recently been shown to offer advantages in both accuracy and predictive ability over traditional statistical methods. However, model development is complicated by the fact that back-propagation neural networks contain several adjustable parameters whose optimal values are initially unknown. These include learning rate, initial weight range, momentum, and training tolerance, as well as the network architecture. The effect of these factors on network performance is investigated here by means of a D-optimal experiment. The goal is to determine how the factors impact network performance and to derive a set of parameters which optimize performance based on several criteria. The network responses optimized are learning capability, predictive capability, and training time. Learning and prediction accuracy are quantified by the experimental error of the model. The process modeled is polysilicon etching in a CCl ₄-based plasma. Statistical analysis of the experimental results reveals that learning capability and convergence speed depend mostly on the learning parameters, whereas prediction is controlled primarily by the number of hidden layer neurons. An optimal network structure and parameter set has been determined which minimizes learning error, prediction error, and training time individually as well as collectively     

A dynamic learning neural network for remote sensing applications

The neural network learning process is to adjust the network weights to adapt the selected training data. Based on the polynomial basis function (PBF) modeled neural network that is a modified multilayer perceptrons (MLP) network, a dynamic learning algorithm (DL) is proposed. The presented learning algorithm makes use of the Kalman filtering technique to update the network weights, in the sense that the stochastic characteristics of incoming data sets are implicitly incorporated into the network. The Kalman gains which represent the learning rates of the network weights updating are calculated by using the U-D factorization. By concatenating all of the network weights at each layer to form a long vector such that it can be updated without propagating back, the proposed algorithm improves the performance of convergence to which the backpropagation (BP) learning algorithm often suffers. Numerical illustrations are carried out using two categories of problems: multispectral imagery classification and surface parameters inversion. Results indicates the use of Kalman filtering algorithm not only substantially increases the convergence rate in the learning stage, but also enhances the separability for highly nonlinear boundaries problems, as compared to BP algorithm, suggesting that the proposed DL neural network provides a practical and potential tool for remote sensing applications     

Network traffic prediction method based on improved ABC algorithm optimized EM-ELM

In order to overcome the poor generalization ability and low accuracy of traditional network traffic prediction methods, a prediction method based on improved artificial bee colony (ABC) algorithm optimized error minimized extreme learning machine (EM-ELM) is proposed. EM-ELM has good generalization ability. But many useless neurons in EM-ELM have little influences on the final network output, and reduce the efficiency of the algorithm. Based on the EM-ELM, an improved ABC algorithm is introduced to optimize the parameters of the hidden layer nodes, decrease the number of useless neurons. Network complexity is reduced. The efficiency of the algorithm is improved. The stability and convergence property of the proposed prediction method are proved. The proposed prediction method is used in the prediction of network traffic. In the simulation, the actual collected network traffic is used as the research object. Compared with other prediction methods, the simulation results show that the proposed prediction method reduces the training time of the prediction model, decreases the number of hidden layer nodes. The proposed prediction method has higher prediction accuracy and reliable performance. At the same time, the performance indicators are improved.     

一个新的二进前向多层网学习算法及布尔函数优化实现

本文首先给出二进前向多层网几何学习算法的一个改进策略,提高了原算法的学习效率,然后同个新的神经网络启发式遗传几何学习算法。ＨＧＧＬ算法采用面向知识的交叉算子和变异算子对几何超平面进行优化的划分,同时确定隐层神经元的个数及连接权系数和阈值,对任意布尔函数,ＨＧＧＬ算法可获得迄今为止隐节点数量少的神经网络结构。     

Accelerated training of backpropagation networks by using adaptive momentum step

Considerable research into the training of neural networks by the backpropagation technique has been undertaken in recent years. Introduction of a momentum term into the training equation can accelerate the training process. A new momentum step and a scheme for dynamically selecting the momentum rate are described. It is shown that these give improved acceleration of training and strong global convergence characteristics. Results are presented for four benchmark training tasks.<>     

复数前馈神经网络的全局最优和快速学习算法

提出了一种新型复数前馈神经网络的学习算法。当输入层和隐层之间的权值计算出来后，就可以通过求解线性方程组得到隐层和输出层之间的权值。这些权值是全局最小点。另一方面，本文算法很容易确定全局最小点时隐层神经元的个数。本文算法具有很高的训练精度和学习速度。     

FPGA-implementation of dynamic time delay neural network for power amplifier behavioral modeling

In this paper, we propose new architectures for FPGA-implementation of a dynamic neural network power amplifier behavioral modeling. The real-valued time-delay neural network (RVTDNN) and the backpropagation (BP) learning algorithm were implemented on FPGA using Xilinx system generator for DSP and the Virtex-6 FPGA ML605 evaluation kit. Different RVTDNN architectures are proposed for various values of the number of hidden neurons, the activation function resolution, and the fixed-point data format. These architectures are evaluated and compared in terms of modeling performances and resource utilization using 16-QAM modulated test signal.     

A neural network-based tracking control system

An application of the backpropagation neural network to the tracking control of industrial drive systems is presented. The merits of the approach lie in the simplicity of the scheme and its practicality for real-time control. Feedback error trajectories, rather than desired and/or actual trajectories, are employed as inputs to the neural network tracking controller. It can follow any arbitrarily prescribed trajectory even when the desired trajectory is changed to that not used in the training. Simulation was performed to demonstrate the feasibility and effectiveness of the proposed scheme     

Temperature control with a neural fuzzy inference network

Although multilayered backpropagation neural networks (BPNNs) have demonstrated high potential in adaptive control, their long training time usually discourages their applications in industry. Moreover, when they are trained online to adapt to plant variations, the over-tuned phenomenon usually occurs. To overcome the weakness of the BPNN, we propose a neural fuzzy inference network (NFIN) suitable for adaptive control of practical plant systems in general and for adaptive temperature control of a water bath system in particular. The NFIN is inherently a modified Takagi-Sugeno-Kang (TSK)-type fuzzy rule based model possessing a neural network's learning ability. In contrast to the general adaptive neural fuzzy networks, where the rules should be decided in advance before parameter learning is performed, there are no rules initially in the NFIN. The rules in the NFIN are created and adapted as online learning proceeds via simultaneous structure and parameter identification. The NFIN has been applied to a practical water bath temperature control system. As compared to the BPNN under the same training procedure, the simulated results show that not only can the NFIN greatly reduce the training time and avoid the over-tuned phenomenon, but the NFIN also has perfect regulation ability. The performance of the NFIN is compared to that of the traditional PID controller and fuzzy logic controller (FLC) on the water bath temperature control system. The three control schemes are compared, with respect to set point regulation, ramp-point tracking, and the influence of unknown impulse noise and large parameter variation in the temperature control system. The proposed NFIN scheme has the best control performance     

Speed control of ultrasonic motors using neural network

The ultrasonic motor (USM) is a newly developed motor, and it has excellent performance and many useful features, therefore, it has been expected to be of practical use. However, the driving principle of USM is different from that of other electromagnetic-type motors, and the mathematical model is complex to apply to motor control. Furthermore, the speed characteristics of the motor have heavy nonlinearity and vary with driving conditions. Hence, the precise speed control of USM is generally difficult. This paper proposes a new speed-control scheme for USM using a neural network. The proposed controller can approximate the nonlinear input-output mappings of the motor using a neural network and can compensate the characteristic variations by on-line learning using the error backpropagation algorithm. Then, the trained network finally makes an inverse model of the motor. The usefulness and validity of the proposed control scheme are examined in experiments