Communication-Driven Learning-Based Harmonic Mitigation Approach for Grid-Forming Converters

Abstract

This paper proposes a harmonic mitigation framework for multi-bus microgrids (MGs) that utilizes a communication-driven learning-based approach. A dedicated neural network is used to model additional intelligent control loops. We also develop a global harmonic distortion (GHD) matrix, formulated by the total harmonic distortion at each point of common coupling (PCC). The GHD specifically focuses on harmonic orders that exceed acceptable limits observed from multi-bus systems. The GHD is used as an input signal of the neural network. Based on the system conditions, the developed GHD allows the system to intelligently trigger the control signal whenever dominant harmonic orders are detected and need mitigation. Other input signals include the voltage and current vectors from all PCCs, which contain local harmonics. To train the network, various critical system operating points are collected as time-series data. At this stage, system parameters are not needed; however, secure communication for data transfer is considered instead. Consequently, the collected data are used to train the network using the developed loss functions that focus on harmonic mitigation. After network training, the output signal is sent to the summing points in the direct axis of the inner voltage and current control loops of different models of grid-forming IBRs units. Simulation results are verified in a modified multi-bus MG with IBRs under various operating conditions.