June 28, 1998
June 28, 1998
July 1, 1998
3.422.1 - 3.422.5
Neural Network Adaptive Autotuner Alireza Rahrooh, Bahman Motlagh University of Central Florida
Abstract It is critical that modern control theory techniques be integrated into assignments which involve the application of basic concepts in engineering technology to prepare students for the next millennium. The adaptive neural network discussed in this paper can be viewed as an appropriate use of these modern techniques in engineering technology curriculum. Adaptive tuning of PID controller gains in case of plant parameter variations is of great importance. There are many approaches available for PID autotuning. In this paper the PID controller gains are adaptively changed using a neural network approach. The neural network tuner is incorporated in the control system to adapt the PID gains to changing system parameters. The neural network architecture employed is a multilayer perceptron. A computer simulation is conducted to show the tracking behavior of the controller in the case of plant parameter variations and set point changes.
Proportional-Integral-Derivative (PID) controllers are the most widely used controllers in the industry. This wide applicability is due to their simple structure and robust behavior. The proper choice of the PID gains is essential for a satisfactory performance. Many attempts for tuning these 1-3 controllers are mainly achieved by off-line tests performed on the plant . There are also other more 4-7 modem autotuners available . The controller parameters, regardless of their method of tuning, must be retuned after a change of plant parameters due to changing operating points, aging, etc. Therefore, the design of an autotuner that adaptively changes the controller gains could be of practical interest.
In this paper, neural networks are utilized to design a tuner that changes the PID gains adaptively. 8-10 Neutral networks have been previously used as adaptive controllers to control unknown plants . The main idea developed in this paper is to use neural networks as adaptive autotuners rather than as an adaptive controller. Simulation results are presented to show the performance of this autotuner in the case of plant parameter variations and set point changes.
The Neural Network Structure
A neural network which is widely used in the control applications is the feedforward network with 11 two hidden layers. The learning algorithm used in the present network is backpropagation . A problem with this network is its rather long learning time. In order to reduce this time, a linear neuron is used in the outer layer and sigmoid neurons are used in hidden layers, respectively. Figure 1 shows such a network (S is sigmoid neuron, and L is linear neuron). The required learning time is considerably reduced by accelerating the convergence rate and reducing the initial value of the norm of the error vector. To further improve the performance of the network the momentum algorithm is
Motlagh, B., & Rahrooh, A. (1998, June), Neural Network Adaptive Autotuner Paper presented at 1998 Annual Conference, Seattle, Washington. https://peer.asee.org/7307
ASEE holds the copyright on this document. It may be read by the public free of charge. Authors may archive their work on personal websites or in institutional repositories with the following citation: © 1998 American Society for Engineering Education. Other scholars may excerpt or quote from these materials with the same citation. When excerpting or quoting from Conference Proceedings, authors should, in addition to noting the ASEE copyright, list all the original authors and their institutions and name the host city of the conference. - Last updated April 1, 2015