Moment Projection Feedforward Control Adaptation

The last several years has seen the commercialization of the Pattern Recognition approach to adaptive control, This technique has several basic advantages: ¿ Performance feedback to a specified performance shape. ¿ Independence from any particular assumption of modeling form. ¿ Insensitive to mismodeling. ¿ Insensitive to reasonable process nonlinearities. ¿ Tuning based on total response behavior ¿ Compatible with the well-understood PID Controller. ¿ Compatible with intuitive operational understanding. ¿ Naturally lends itself to arbitrary refinement as an expert system. It has been argued that, while the Pattern Recognition technique is good enough for single-loop control, something more elaborate will be needed for multivariable control: either a model-based system or some very sophisticated expert system approach. Without conceding the point totally, model-based techniques are certainly convenient for multivariable control adaptation. This paper presents a unique, patent pending, multivariable/feedforward adaptive technique, designed to work well with single-loop Pattern Recognition Adaptation. It recognizes certain nuances of adaptation technology, when applied to feedforward systems, which allow rigorous model-based adaptation not possible for feedback control. It also recognizes possibilities for insensitive multiparameter updating and effective "forgetting" when transients are adapted as a unit, not possible under conventional model-based designs. The new method combines two methods generally held in ill-favor in the literature: the Method of Moments and the Projection Method. The result is a method having the modeling insensitivity and intuitive behavior of the Pattern Recognition Adaptation, plus several advantages of its own, significant to multivariable control: ¿ Structure supporting block diagram modular application.