This paper elucidates the working of shunt active filters in dynamically varying grid conditions. The accuracy of harmonics eliminated is based on the harmonic extraction technique and the switching techniques used. The harmonics of the electrical system are computed and based on the various harmonics produced; the shunt active filter has to supply a compensating current in opposition at the point of common coupling. Space Vector Pulse Width modulation (SVPWM) is an effective switching technique used but its performance is not satisfactory in the over-modulation regions. In this paper, optimised switching technique using a neural network for SVPWM is used which has lesser computation time and computes the gating pulses effectively in the over-modulation regions. Shunt active filter mitigates the impact of harmonics even in the over-modulation regions. The hardware is implemented using FPGA Spartan 6 controller, which utilises fewer hardware resources. The optimised switching topology results with improved harmonic reduction even in the over-modulation regions with unbalanced loading conditions is presented in this paper.