Artificial neural networks are mathematical tools inspired by what is known about the physical structure and mechanism of the biological cognition and learning. Neural networks have attracted considerable attention du...
详细信息
Artificial neural networks are mathematical tools inspired by what is known about the physical structure and mechanism of the biological cognition and learning. Neural networks have attracted considerable attention due to their efficacy to model wide spectrum of challenging problems. In this paper, we present one of the most popular networks, the backpropagation, and discuss its learning algorithm and analyze several issues necessary for designing optimal networks that can generalize after being trained on examples. As an application in the area of predictive microbiology, modeling of microorganism growth by neural networks will be presented in a second paper of this series.
Random sample selection method in backpropagation results in convergence on the error (root of mean squared error, RMSE) surface. These problems, which are caused by the extreme (worst-case) errors, can be solved by a...
详细信息
Random sample selection method in backpropagation results in convergence on the error (root of mean squared error, RMSE) surface. These problems, which are caused by the extreme (worst-case) errors, can be solved by a different sample selection strategy. A sample selection strategy has been proposed, which provides lower maximal errors and a higher confidence level on the expense of slightly increased RMSE. Applications are presented in the held of spectroscopic ellipsometry (SE), a sensitive, non-destructive but indirect analytical technique. Demonstrative example shows feature common to simulated annealing in the sense of escaping local minima.
A new design method for spatially-homogeneous, fully recurrent neural networks is presented. In our approach the eigenvalues of the synaptic matrix, rather than the weights, are learned from the examples. When the lea...
详细信息
A new design method for spatially-homogeneous, fully recurrent neural networks is presented. In our approach the eigenvalues of the synaptic matrix, rather than the weights, are learned from the examples. When the learning process is carried out, the connection weights are easily computed from the eigenvalues by inverse discrete Fourier transform. The adaptation is performed in the eigenvalue space in older to simply incorporate in the training algorithm the conditions for the uniqueness of the steady-state. As a consequence, the trained networks are insensitive to initial conditions. The method is illustrated by computer simulations concerning two specific feature extraction examples. Copyright (C) 1996 Elsevier Science Ltd.
In this paper, a nonlinear control strategy based on using a shape-tunable neural network is developed for adaptive control of nonlinear processes. Based on the steepest descent method, a learning algorithm that enabl...
详细信息
In this paper, a nonlinear control strategy based on using a shape-tunable neural network is developed for adaptive control of nonlinear processes. Based on the steepest descent method, a learning algorithm that enables the neural controller to possess the ability of automatic controller output range adjustment is derived. The novel feature of automatic output range adjustment provides the neural controller more flexibility and capability, and therefore the scaling procedure, which is usually unavoidable for the conventional fixed-shape neural controllers, becomes unnecessary. The advantages and effectiveness of the proposed nonlinear control strategy are demonstrated through the challenge problem of controlling an open-loop unstable nonlinear continuous stirred tank reactor (CSTR).
A new adaptive pole placement controller for unknown nonlinear plants is developed using a modified neural network. The modified neural network is composed of two parts: one is a linear neural network (LNN), which is ...
详细信息
ISBN:
(纸本)0780338332
A new adaptive pole placement controller for unknown nonlinear plants is developed using a modified neural network. The modified neural network is composed of two parts: one is a linear neural network (LNN), which is the linearised model at the operating point;and the other is a multilayered feedforward neural network (MFNN), which approximates the nonlinear dynamics of the plant that can not be modelled by the LNN. Then a fast learning algorithm is presented for training the proposed neural network. The controller design is based on the LNN, and the output of the MFNN is considered as a measurable disturbance and is eliminated through feedforward control. Simulation results reveal that the proposed training algorithm is much faster than the standard algorithm and the new adaptive pole placement controller can effectively control a class of nonlinear plants.
This paper presents an adaptive PID-like controller(PIDLC) using a modified Neural network(MNN) for learning the characteristics of a dynamic system. The PIDLC can adapt parameters' variation and uncertainty in th...
详细信息
ISBN:
(纸本)0780338324;0780338332
This paper presents an adaptive PID-like controller(PIDLC) using a modified Neural network(MNN) for learning the characteristics of a dynamic system. The PIDLC can adapt parameters' variation and uncertainty in the controlled plant through on-line learning. The MNN's learning algorithm is considerably faster because of the introduction of recursive least squares(RLS) algorithm. The simulation results show that this kind of control algorithm is very effective especially when there are variations in the plant dynamics.
This paper presents a minimum variance predictive controller (MVPC) using a modified Neural network(MNN) in order to learn the characteristics of a dynamic system. The MVPC can adapt parameters' variation and unce...
详细信息
ISBN:
(纸本)0780342534
This paper presents a minimum variance predictive controller (MVPC) using a modified Neural network(MNN) in order to learn the characteristics of a dynamic system. The MVPC can adapt parameters' variation and uncertainty in the controlled plant through the on-line learning. The learning algorithm is considerably faster because of the introduction of recursive least squares(RLS) algorithm. Simulation results have shown that the proposed approach is effective for adaptive control of nonlinear systems.
A new adaptive pole placement controller for nonlinear systems using a modified neural network is presented. The modified neural network is composed of two parts: one is a linear neural network (LNN), and the other is...
详细信息
A new adaptive pole placement controller for nonlinear systems using a modified neural network is presented. The modified neural network is composed of two parts: one is a linear neural network (LNN), and the other is a multilayer feedforward neural network C:MFNN). Then a fast learning algorithm is proposed for training the network. The adaptive control design is based on the LNN and MFNN. Simulation results reveal that the new adaptive pole placement controller can effectively control a class of nonlinear systems. Un neuf adaptable pôle assignation contrôleur pour non-linéaire systèms lequels fait usage de modifiable nerveux réseau est présenté ici. Le modifiable nerveux réseau est compose de deux parts: un linéaire nerveux réseau (LNR) et un non-linéaire mutilassise nerveux réseau (NMNR). Une méthode pour apprendre algorithme rapidement est offert aussi. Le adaptable contrôl dessein est basé sur LNR et NMNR. Les simulation résultats révélent que le neuf adaptable pôle assignation contrôleur peut contrôller effectivement un type des non-linélaire systèmes.
Minsky and Papert (Perceptons: An Introduction to Computational Geometry, MIT Press: Cambridge, AA, 1969) show that a two-layer perceptron with monotonic activation function cannot solve the Xor problem. We present a ...
详细信息
Minsky and Papert (Perceptons: An Introduction to Computational Geometry, MIT Press: Cambridge, AA, 1969) show that a two-layer perceptron with monotonic activation function cannot solve the Xor problem. We present a two-layer perceptron with non-monotonic activation function which can separate non-linearly separated sets of data. This kind of perceptron is then applied to the Xor problem, parity problem and a pattern recognition problem. learning strategy is detailed, and performance is evaluated.
A new on-line direct control scheme for the Autonomous Underwater Vehicles (AUV), using recurrent neural networks, is investigated. In the proposed scheme, the controller consists of a three-layer network architecture...
详细信息
A new on-line direct control scheme for the Autonomous Underwater Vehicles (AUV), using recurrent neural networks, is investigated. In the proposed scheme, the controller consists of a three-layer network architecture having feedforward input and output layers, and a totally recurrent hidden layer All the interconnection strengths are synchronously updated using a computationally inexpensive learning algorithm called Alopex. The updating is based on the output error of the system directly, rather than using a transformed version of the error employed in the other neural network based direct control schemes. In the present implementation, the network starts from random initial conditions without needing any prior training, and learns the dynamics of the AUV to provide the correct control signal. Based on the simulation experiments using the nonlinear dynamics of an AUV, we demonstrate that the proposed learning algorithm and the network architecture provide stable and accurate tracking performance. We have also addressed the issue of robustness of the controller to system parameter variations as well as to measurement disturbances.
暂无评论