High-speed automated identification and design of physical systems using hybrid computer optimization techniques

A method has been developed for high-speed automated identification and design of linear dynamic physical systems and circuits using hybrid computer optimization techniques. The method has been extended to a specified class of nonlinear problems. Cost and speed advantages for hybrid computation over all-digital computation have been found for this class of problems. A technique of model structure optimization, believed to be original, eliminates the usual assumption of model form with the result that the optimum model form as well as the optimum parameter set is found by "growing" models. Specialized hardware developments are required for structure optimization. Results obtained by the use of these techniques and devices are presented.