An Adaptive Secondary Control Based on Recursive Neural Networks for Isolated Microgrids

Abstract

This work proposes a model reference adaptive control based on recursive neural networks. This secondary-level controller corrects the deviations on the voltage and frequency setpoints of a simple primary control in an isolated microgrid (MG) with multiple distributed generators (DGs). The controller has two nonlinear autoregressive with external input neural networks to emulate the microgrid and guide the whole system according to an appropriate reference model. The networks are trained with synthetic data from a simulation MG and the order of the networks are obtained with a deterministic method. The method is tested with a three-units MG in a Matlab simulation under different working conditions. Results show the proper performance of the proposed controllers in comparison to a PI strategy.