Neural network indirect adaptive control with fast learning algorithm

doi:10.1016/0925-2312(95)00091-7

Neurocomputing

Volume 13, Issues 2–4, October 1996, Pages 185-199

https://doi.org/10.1016/0925-2312(95)00091-7 Get rights and content

Abstract

A fast learning algorithm based on a new cost function and a linearized error signal is proposed. The proposed learning algorithm is applied to indirect adaptive control of nonlinear plants. In the proposed method, we use the identification error and the control error to train the NNI and the NNC, respectively. In addition, we introduce a linearized error signal in order to improve the learning speed. Computer simulation results show that the rate of convergence increases, and that the NNC based on the proposed method is insensitive to variations of the plant parameters.

References (8)

K.J. Hunt et al.
Neural networks for control systems — A survey
Automatica
(1992)
D.E. Rumelhart et al.
K.S. Narendra et al.
Gradient methods for the optimization of dynamical systems containing neural networks
IEEE Trans. Neural Networks
(1991)
H.M. Tai et al.
A neural network-based tracking control system
IEEE Trans. Ind. Electron.
(1992)

There are more references available in the full text version of this article.

Cited by (30)

Identification and optimal control of nonlinear systems using recurrent neural networks and reinforcement learning: An overview
2021, Neurocomputing
Citation Excerpt :
Fig. 2 shows the diagram of the discrete recurrent net with hidden layers. Stability analysis of discrete recurrent nets with hidden layers are available [14,15,52]. The past sections address the identification of nonlinear system (1) using recurrent neural networks without (SLP) or with hidden layers (MLP).
This paper reviews the identification and optimal control problems using recurrent neural networks and reinforcement learning for nonlinear systems both in discrete- and continuous-time. Since neural networks can approximate any nonlinear function, then is shown that it can approximate a dynamical system using some well-identified elements and different neural structures. Existing methods using Lyapunov and Riccati equations to get the neural weights update rules are reviewed. Optimal control using a linear quadratic regulator formulation or reinforcement learning methods are discussed. We discuss the normalized gradient descent algorithm as core algorithm for the optimal control design using reinforcement learning.
Neural network control of nonlinear dynamic systems using hybrid algorithm
2014, Applied Soft Computing Journal
Citation Excerpt :
By combining the above two modes, the learning algorithm converge much faster than other well-known algorithms. The main issues in the field of neural networks based control include the architecture of the identifier and controller, and the algorithm to be used in training their parameters [50,51]. The proposed control architecture adopted in this paper is the basis for many current neurocontroller.
In this paper, a hybrid method is proposed to control a nonlinear dynamic system using feedforward neural network. This learning procedure uses different learning algorithm separately. The weights connecting the input and hidden layers are firstly adjusted by a self organized learning procedure, whereas the weights between hidden and output layers are trained by supervised learning algorithm, such as a gradient descent method. A comparison with backpropagation (BP) shows that the new algorithm can considerably reduce network training time.
A multivariable on-line adaptive PID controller using auto-tuning neurons
2003, Engineering Applications of Artificial Intelligence
In this paper, we present a new PID control technique based on auto-tuning neurons for multivariable systems. The main difference between an auto-tuning neuron and a general neuron is that there are adjustable parameters of the activation function used in an auto-tuning neuron. In this paper, a modified hyperbolic tangent function is used as the activation function of an auto-tuning neuron, which provides two adjustable parameters to flexibly determine the magnitude and the shape of function. We then use such auto-tuning neurons to find gains of the multivariable PID controller, which is tuned on-line according to certain adaptation laws. Finally, two illustrative examples will be used to compare the performance by using our proposed method and other methods.
Stable direct adaptive neural controller of nonlinear systems based on single auro-tuning neuron
2002, Neurocomputing
In this paper, the design of a stable direct adaptive neural controller based on single auto-tuning neuron for nonlinear systems is developed. Unlike traditional multi-layered neural controller, the structure of the proposed controller is simpler and practical. Only a single neuron with three adjustable parameters is employed in our proposed neural controller. The adaptation law to adjust these parameters is proposed based on the Lyapunov approach. Moreover, the stability of the system can be also analyzed and guaranteed by introducing the supervisory controller and the modified adaptation law with projection. Such a controller design is very easy for hardware implementation due to its simple structure. Finally, the simulations of an unstable nonlinear system by using the proposed single auto-tuning neural controller will be illustrated to demonstrate the excellent control capability.
Identification and control of a riser-type FCC unit using neural networks
2002, Computers and Chemical Engineering
This paper addresses the use of feedforward neural networks for the steady-state and dynamic identification and control of a riser type fluid catalytic cracking unit (FCCU). A back propagation algorithm with momentum term and adaptive learning rate is used for training the identification networks. The back propagation algorithm is also used for the neuro-control of the process. It is shown that for a noise-free system the adaptive neuro-controller is capable of maintaining the riser temperature, the pressure difference between the reactor vessel and the regenerator, and the catalyst bed level in the reactor vessel, in the presence of set-point and disturbance changes.
A parallel neuro-controller for DC motors containing nonlinear friction
2000, Neurocomputing
This paper presents an application of a parallel neuro-controller for compensating the effects induced by the friction in a DC motor system. A back-propagation neural network based on a gradient descent algorithm is employed, and a bound on the tracking error is derived from the analysis of the tracking error dynamics. The parallel neuro-controller is a combination of a linear controller and a neural network controller which compensates for nonlinear friction. The proposed scheme is implemented and tested on an IBM PC-based DC motor control system. The algorithm, simulations, and experimental results are described. The results are relevant for many precision drives, such as those found in industrial robots.

View all citing articles on Scopus

View full text

Special paperNeural network indirect adaptive control with fast learning algorithm

Abstract

Automatica

Gradient methods for the optimization of dynamical systems containing neural networks

IEEE Trans. Neural Networks

A neural network-based tracking control system

IEEE Trans. Ind. Electron.

Special paper
Neural network indirect adaptive control with fast learning algorithm