Rovithakis G.A., Christodoulou M.A. Adaptive Control with Recurrent High-order Neural Networks. Theory and Industrial Applications

Springer, 2000. — 203 p.Recent technological developments have forced control engineers to deal with extremely complex systems that include uncertain, and possibly unknown, nonlinearities, operating in highly uncertain environments. The above, together with continuously demanding performance requirements, place control engineering as one of the most challenging technological fields. In this perspective, many "conventional" control schemes fail to provide solid design procedures, since they mainly require known mathematical models of the system and/or make assumptions that are often violated in real world applications. This is the reason why a lot of research activity has been concentrated on "intelligent" techniques recently.
One of the most significant tools that serve in this direction, is the so called artificial neural networks (ANN). Inspired by biological neuronal systems, ANNs have presented superb learning, adaptation, classification and function approximation properties, making their use in on line system identification and closed-loop control promising.
Early enrolment of ANNs in control exhibit a vast number of papers proposing different topologies and solving various application problems. Unfortunately, only computer simulations were provided at that time, indicating good performance. Before hitting real-world applications, certain properties like stability, convergence and robustness of the ANN-based control architectures, must be obtained although such theoretical investigations though started to appear no earlier than 1992.
The primary purpose of this book is to present a set of techniques, which would allow the design of controllers able to guarantee stability, convergence and robustness for dynamical systems with unknown nonlinearities, real time schedulers for manufacturing systems.
To compensate for the significant amount of uncertainty in system structure, a recently developed neural network model, named Recurrent High Order Neural Network (RHONN), is employed. This is the major novelty of this book, when compared with others in the field. The relation between neural and adaptive control is also clearly revealed.
It is assumed that the reader is familiar with a standard undergraduate background in control theory, as well as with stability and robustness concepts. The book is the outcome of the recent research efforts of its authors. Although it is intended to be a research monograph, the book is also useful for an industrial audience, where the interest is mainly on implementation rather than analyzing the stability and robustness of the control algorithms. Tables are used to summarize the control schemes presented herein.Identification of Dynamical Systems Using Recurrent High-order Neural Networks
Indirect Adaptive Control
Direct Adaptive Control
Manufacturing Systems Scheduling
Scheduling using RHONNs: A Test Case