A Mathematical Programming Approach to Identification and Optimization of Complex Systems,
Abstract
Complex systems are often represented using models which are not traditional, closed-form mathematical expressions. Such a model is essentially unknown from the standpoint of a mathematical programming algorithm because the information which is important to the algorithm is only available through experimentation. Extension of mathematical programming to complex systems in which the cost of information is significant leads to combining system identification with optimization. Inner linearization result in computational approaches which combine identification and optimization. The pricing problem associated with restriction when inner linearization is applied selectively is derived using the Kuhn-Tucker conditions. Three approaches to the identification-optimization problem are proposed: inner linearization followed by restriction (ILR); outer linearization followed by relaxation (OLR); and a sequential combination of inner- and outer-linearized subproblems (SIO). Computational experience showed that (ILR) is dominated by (OLR) and (SIO), and that (SIO) is slightly better than (OLR) for some problems. A 'natural' termination rule was evaluated and shown to be significantly inferior to an ex post facto 'optimal' rule. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1970
- Accession Number
- AD0714832
Entities
People
- Charles A. Holloway
Organizations
- University of California, Los Angeles