Date of Award
Dissertation - Restricted
Doctor of Philosophy (PhD)
Electrical and Computer Engineering
The focus of this work is on the development and utilization of artificial neural networks (ANNs) for the purpose of providing improved identification of plants exhibiting significant nonlinearity and uncertainty. The approach is to integrate ANNs into parallel-series identification structures as general mappings capable of representing the unknown portion of a plant. The problem is then divided between the realization of suitable identification structures and the development of adequate learning algorithms. The contributions of this research are as follows. First, advanced learning algorithms are explored for the purpose of parameterizing ANN identification structures. A consequence of this investigation is the development of an original ANN training algorithm that exhibits superior identification in three specific applications when compared to four other previously reported techniques. The ANN framework is improved through the development of a novel method of exploiting a priori information about a given plant through the realization of a new class of ANN layers. The methods are used to identify the static torque curves of a switched reluctance motor and are demonstrated through four different examples involving the identification of nonlinear plants. Finally, an on-line parallel-series identification model is proposed that utilizes ANNs and the methods developed in earlier chapters to identify a SRM and its load characteristics. The model is demonstrated through simulation experiments involving a short calibration period comprised of a point-to-point move.