Neural networks and feedback linearization
Khosrow M. Hassibi, Kenneth A. Loparo · 2002
The authors propose a general method for learning of the input-output feedback linearization (IOFL) laws for the class of nonlinear systems described by F.-C. Chen (1990). The direct method previously described by the author (1991) is used with some modifications in the implementation. Three general assumptions required for successful implementation of the method are given. The objective was to learn a controller using a high-order three-layer network such that the resulting closed-loop system behaves similarly to a linear reference model. The IOFL problem is classified into three cases, and the required assumptions to learn the feedback for each case are given. In all cases, the feedback linearizable control law is learned directly from the error between the closed-loop system and the reference model outputs.>