A Data-Driven Approach for Grid Synchronization Based on Deep Learning

Abstract

Synchronization is a complex problem, mainly due to its nonlinearity and stochastic nature of the grid. The Phase-Locked Loop (PLL) has been the standard scheme for the synchronization and P/Q decoupling in grid-following inverters. Nonetheless, during transients, the PLL response is not ideal and causes oscillations or overshoots. Moreover, in adverse grid conditions, the PLL performance degrades significantly and loss of synchronism might occur. This paper introduces a structurally new scheme based on deep neural networks for synchronization, called DeepSynch. The method is capable of extracting the voltage phase fast and in a stable manner, even in a harmonic-polluted environment. The simulation results verify the performance of the proposed scheme.