Generalised neural network-based control algorithm for DSTATCOM in distribution systems

This study presents a new concept to control a distribution static compensator (DSTATCOM) based on generalised neural network in a three-phase power distribution system. Artificial neural network (ANN)-based controllers play the vital role in the performance improvement of DSTATCOM. However, their application is limited by the increase in complexity as well as the computational time. The proposed generalised neural network algorithm is a combination of Gaussian, sigmoidal and linear transfer functions within a layer to improve the DSTATCOM control strategy. This algorithm estimates the amplitude of the wattful and wattless current components of the load currents for harmonics elimination and reactive power compensation by the DSTATCOM. The algorithm is developed in MATLAB. The case studies validate its superiority over ANN-based control algorithms. The proposed method needs a less number of training patterns and unknown weights compared to other algorithms which reduces the complexity and the computational time. It also improves the performance of DSTATCOM estimating the weights and its learning online which is of the main merits of this algorithm. Its other inherent advantages are ease in design, robustness and its adaptivity with dynamics of load at utility end.