Self-Organizing Artificial Neural Networks in Dynamic Systems Modeling andControl

摘要

This thesis discusses the potential of self-organizing networks applied to themodeling and control of dynamic systems. A general framework is created for self-organizing networks applications that is especially suited to control engineering. Apart from the traditional time-domain features, linear operators can be mapped in the network. Efficient algorithms are derived for adapting the parameters in the self-organizing net. These algorithms are motivated as solutions to well-founded optimization problems. Two prototypical approaches to realizing controllers based on the neural network representations are presented with simple application examples. The theoretical problems of combining self-organizing networks with dynamic systems are discussed extensively. The algorithms that are derived for self-organization are not limited only to control engineering applications. On the other hand, the theoretical analyses of dynamic systems are not limited to neural networks applications--for example, the identification algorithms and the results achieved on the systems identifiability are applicable to the general parameter estimation tasks.