MINIMAX CONTROLLER DESIGN,
Abstract
When an uncertain dynamic system is to be controlled in an optimal manner, that is when the controller is required to minimize a performance index, the controller design requires a compromise between controller complexity and system performance. At one extreme is the optimal-adaptive controller which is difficult to realize but yields ideal performance, and at the other is any overly simplified controller which yields unacceptable performance. The reasonable controller structures for a given system can often be determined by the designer in terms of a number of free parameters. Then for each structure it is desired to find those parameters which yield the 'best' system performance. This thesis develops minimax methods for determining such controller parameters. The concept of performance-sensitivity is introduced to meet the usual criticism of minimax or 'worst-case' design, that it is too pessimistic in concentrating all attention on the worst parameters. Properties of minimax control with a performance-sensitivity as the index are developed. It is shown that the usually desired range of system properties can be achieved by minimaximizing either the system performance index or a performance-sensitivity. A new algorithm for solving algebraic minimax problems, regardless of the presence of a saddle point, is presented and proved to converge. The rate of convergence, and simplifications which occur when the system is linear or when the index has convexity properties, are discussed. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1967
- Accession Number
- AD0656450
Entities
People
- David M. Salmon
Organizations
- University of Illinois Urbana–Champaign