Shunt active filtering with NARX feedback neural networks based reference current generation

This paper proposes a nonlinear autoregressive exogenous (NARX) feedback neural networks (NN) based reference current generation (RCG) scheme for 3-phase shunt active filter (SAF). NARX feedback NN is employed to implement the frequency independent RCG scheme. Usually, such schemes need to be individually implemented for each phase resulting in increased complexity. Alternately, NARX feedback NN processes the 3-phase current quantities and unit vector templates (UVT) for the estimation of fundamental active component of load current, compensating currents and consequently, reference source currents. The inputs for NARX feedback NN are the previous two sample of the estimated compensating current and fundamental active component of the load current, along with the present and the previous samples of load current and UVT. The control system ensures that the source currents match the respective reference values resulting in total harmonic distortion less than 5% and unity power factor at the supply end. Thus, with the proposed scheme, the SAF eliminates current harmonics, and compensates for the reactive power and load unbalancing. The performance of SAF with NARX feedback NN based RCG is analyzed under load variations, frequency variations, load unbalancing and distorted supply.