Determination of the control gains of a fuzzy PID controller using neural networks

In this paper a parameter estimation method using multilayer neural networks is developed for finding the control gains of a fuzzy-PID controller. In the F-PID controller developed by the authors (1996), the fuzzy control gains were tuned manually until acceptable gains that produced desired plant output were obtained. In this work, this process is automated by training a neural network to learn and find the control gains of the F-PID controller, given the desired system output. The controller consists of: 1) a multilayer neural network that estimates the control gains of the F-PID controller; and 2) the F-PID controller that uses the gains obtained from the neural network to control a process (plant). In addition, the neural network employs an algorithm that reduces the number of hidden neurons to a minimum. The F-PID control gains obtained from the neural network are tested on higher-order, time-delayed linear and nonlinear plants.