Successive Refinement of Empirical Models

An approach for automatic control using on-line learning is presented. The controller’s model for the system represents the effects of control actions in v,’u’ious (hyperspherical) regions of state space. The partition produced by these regions is initially coarse, limiting optimization of the corresponding control actions. Regions contract periodically to enable optimization to progress further. New regions and associated optimizing elements are generated to maintain coverage of state space, with characteristics drawn from neighboring regions. During on-line operation the resulting state space partition undergoes successive refinement, with "generalization" occurring over successively smaller areas. The system model effectively grows and becomes more accurate with time.