A machine learning approach to intelligent adaptive control

An attempt to define intelligent adaptive control is presented. It is noted that it might denote a more flexible redesign capability of the control system. For example, suppose the control engineer himself remains in the 'outer' loop. After observing deficiencies in his first attempt at a control system, he might design a replacement that better reflects the eccentricities of the underlying process to be controlled. This research is a first step at automating such a controller. The system itself conjectures a refined system identification and develops a new control algorithm when the previous control system performs poorly. It is pointed out that the research, while promising, is very ambitious and far from complete. It is offered here as a new direction rather than as a mature method for control system design. Planning to achieve different speeds in a simplified single-gear manual transmission automobile is considered by way of illustration of the proposed approach.<
>