A New Adaptive Neural Harmonic Compensator for Inverter Fed Distributed Generation

This paper deals with the command of inverters in DG (distributed generation) systems by use of linear neural networks in such a way that, with a slight upgrade of their control software, they can be used also to compensate for the harmonic distortion in the node where they are connected (local compensation), that is in the in the point of common coupling (PCC). To this purpose a neural estimator based on linear neurons (ADALINEs) has been developed which is able to act as a selective noise cancellers for each harmonic of the node voltage. The use of linear neurons permits the drawbacks of classical neural networks to be overcome and moreover the neural estimator is easy to implement, thus allowing the same inverter to be used also for active filtering. The paper presents and discusses the results obtained by simulation of a small size distribution network, which in itself contains the distorted voltages coming both from the utility and from a nonlinear load and an inverter supplied by an array of photovoltaic panels with its DC/DC internal converter.