Neurocontrol II: High precision control achieved using approximate inverse dynamics models

Cs. Szepesvári and A. Lõrincz

Neural Network World 6, 897--920 (1996)


Abstract


It is common that artificial neural networks (ANNs) are used for approximating the inverse dynamics of a plant. In the accompanying paper a self-organising ANN model for associative identification of the inverse dynamics was introduced. Here we propose the use of approximate inverse dynamic models for both Static and Dynamic State (SDS) feedback control. This compound controller is capable of high-precision control even when the inverse dynamics is just qualitatively modeled or the plant's dynamics is perturbed. Properties of the SDS Feedback Controller in learning the inverse dynamics as well as comparisons with other methods are discussed. An example is presented when a chaotic plant, a bioreactor, is controlled using the SDS Controller. We found that the SDS Controller can compensate model mismatches that otherwise would lead to an untolerably large error if a traditional controller were used.


 * * *

<-- Home