Growing model algorithm for process identification based on neural-gas learning and local linear mapping

Abstract

The paper proposes a special growing type of identification model, based on local linear mapping that gradually improves its accuracy and generalization ability by automatically increasing the size of the model. At each iteration, the feedback information from the approximation error of the current model is utilized in order to make decision for insertion of new local models (units) in the input area with the biggest error. Detailed simulation results and comparisons in the paper have shown that the final produced growing model has a better approximation and generalization ability than some other known learning algorithms. In addition, the proposed procedure automatically defines the optimal size of the model.