CONTACTOR CONTROL BY ADAPTIVE PATTERN-RECOGNITION TECHNIQUES,
Abstract
The purpose of the research described herein is to explore the possibility of utilizing patternrecognition techniques in the design of controllers for complex dynamic processes. In this design reverse-time trajectories are used to generate patterns that indicate the position of the minimum-time optimal switching surface. Adaptive pattern-recognition techniques are used to find a readily implemented switching function that closely approximates the optimal switching surface. An adaptive threshold logic element-a trainable, linear, pattern-recognition device- is used in conjunction with other equipment to realize a trainable, nonlinear, patternrecognition device. The controller-design procedure has been simplified by the ability of the trainable function generator to generalize, i.e., to learn the switching function for all regions of the state space after being trained on patterns from only a part of the space. Conditions have been developed, which, if satisfied by the optimal switching surface, will assure that generalization will occur. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1964
- Accession Number
- AD0443083
Entities
People
- Fred W. Smith
Organizations
- Stanford University