Indirect adaptive control of periodic time-varying systems using neural networks

Abstract

Indirect adaptive control of a class of systems which perform repetitive operations and which are not necessarily time-invariant is considered. Neural networks are used for this purpose with the error committed in onecycle being used to correct the corresponding parameters in the next cycle. It is shown that the time variations of the plant parameters cannot be captured if every parameter is allowed to vary at the sampling rate. Hence various schemes for unfolding the network partially are proposed. Identification and adaptive control algorithms for such schemes are presented. Simulation results on a simple robot are also provided to illustrate the application of the scheme.