Abstract
The design problem of a control input for system parameter identification has so far been studied extensively. On the other hand, the importance of end-time selection should be emphasized in a case where the identification experiment is desired to complete quickly.
In the present paper, at each stage of the experiment, allowed is the decision as to whether it is to be stopped, or continued for further measurement by taking some control. This is formulated as a combined optimal problem of control and stopping by adopting a performance index that has a cost term representing the time required for the experiment. Since the optimal design of the experiment needs the optimum cost function of a number of variables, an attempt is made on simplification of the function into such a form that contains a table-type function of fewer variables and a large amount of information about the optimality equation. Utilizing this form, a learning procedure is proposed to obtain a well-trained stopping rule of interest.
By a numerical example, the suboptimal design thus obtained is examined to see its applicability. The results of computer simulation show that it reduces an appreciable amount of cost and improves the rate of correct estimate of uncertain parameter, as compared with the design having a fixed end-time of the experiment.
