Abstract
Recently, a number of methods on neural network control have been studied. Universal Learning Networks (ULNs) have been proposed, as the name indicates, to provide a universal framework for the class of neural networks and moreover to model and control complex systems. In addition to calculation of the first order derivatives of the signals flowing in the networks that are necessary in gradient-based learning, the generalized ULN learning algorithm is equipped with a systematic mechanism that calculates their second or higher order derivatives. Using the higher order derivatives, robust controlled systems have been studied, that are not much affected by changes in the plant parameters. On the other hand, in the control fields it is requested to design the systems to meet the plural specifications concurrently such as quick response, quick damping and small steady state errors. Until now, there have been thoroughly discussed on how to make a controller to deal with such plural specifications. Model Matching and Model Reference Control system are the typical conventional methods for aiming at such a goal. But, these method have been mainly applied to linear control systems. In this paper, a new control method formulated by ULNs has been proposed, which can overcome the conventional nonlinear problems by using the impulse response as the extended criterion function. In addition the impulse response can be easily calculated by the higher order derivatives of ULNs. In simulations of a nonlinear crane control system, it is shown that the proposed method is superior to the commonly used neural networks control.
