Fractional order hold tuning using neural networks

A connectionist method for autotuning the free parameter of a Fractional order hold (FROH) in order to improve the stability properties of the resulting discrete-time zeros is proposed. Such a technique employs multilayer perceptrons to approximate the mapping between the sampling period/continuous-time parameters of the plant and the optimal values of the FROH parameter. The neural networks are designed to adapt on-line to changing system structures, parameter values and sampling periods. To achieve this objective, the network weighting coefficients are determined during an off-line training phase. In this training phase, the optimal values of the FROH parameter are obtained by applying the classical generalised root locus procedure. Simulation results are presented to illustrate the properties of the complete regression system