The development and performance of a self-optimizing system

Methods are presented for the analysis of conventional hill-climbing optimizing systems. It is shown that, in these systems, rapid optimization is inevitably accompanied by large fluctuations of parameter setting, during both the transient approach to optimum parameter value and operation at this value. By the introduction of a simple sample-and-hold or clamping arrangement which maintains the parameter setting constant over a defined period, the steady-state fluctuations about the optimum can be minimized, and the setting of the parameter approaches the optimum value from any initial off-set in a satisfactory manner. In the proposed system, the fundamental limitation on the speed of optimization is set by measurement errors due to short observation time. The effect of these errors is considered and the resulting requirements on the design of the optimizing loop are discussed. Experimental results from a particular system agree with theoretical predictions, and the responses of a system with two automatically-adjusted parameters, with and without the parameter clamp, are compared.