Adaptive Nonlinear Modeling with Quantizers

摘要

Identification of the dynamics of a nonlinear plant was studied using the combination of a generalized, or equation-error, modeler with multilevel quantizing for piecewise-linear modeling. Algorithms for adapting the quantizing boundary locations and the modeling weights are derived. Conditions are established which allow converting a 'least mean square' error-correcting training algorithm for the modeling weights to a modified forced-learning algorithm which converges as rapidly as the signal statistics permit. Experimental results indicate that this type of modeler converges more rapidly than the ordinary types of nonlinear modelers, and to a more accurate solution. (Author)