HIGH-SPEED EXTREMUM REGULATOR WORKING PER METHOD OF RANDOM SEARCH,
Abstract
An extremal self-adaptive digital regulator for the optimization of objects with many parameters is described. The main part of the regulator is a generator of random pulse sequences, and the optimal mode is determined by random search; this is shown to afford a faster degree of convergence in the adjustment of many-parameter objects than other methods (e.g., the gradient method) and to result in an operating speed two orders of magnitude higher than hitherto developed systems. The extremal regulator is of the 'learning' type, in that parameter changes which lead to improved quality have a greater probability of occurence. The main units of the regulator (random-sequence pulse generator, sequence to amplitude converter (modifier), model of the object, logic block, memory block, limiters) are described in detail, and the results of an experimental check on the operation of the regulator are reported. The adjustment speed increases with increasing size of the discrete steps used in the variation of the parameter and memory. The adjustment accuracy increases with the memory step and decreases with an increasing parameter step. Compensation for null drift and for industrial pickup noise is provided.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 14, 1967
- Accession Number
- AD0662575
Entities
People
- L. A. Rastrigin
- L. V. Sytenko
Organizations
- National Air and Space Intelligence Center