Fault ride-through enhancement for grid-forming virtual synchronous generators in weak grid environments using sequential neural networks
TL;DR
Comparative analysis reveals that the proposed data-driven strategy not only minimizes computational latency but also enhances grid-compliant power quality, achieving a current total harmonic distortion of less than 2%, thereby outperforming the FS-MPC baseline.
Abstract
Grid-forming (GFM) virtual synchronous generators are critical for stabilizing inverter-dominated power grids, yet their low inertia necessitates robust fault ride-through (FRT) and overcurrent protection strategies. While finite-set model predictive control (FS-MPC) provides optimal transient management, its practical deployment is heavily hampered by an exhaustive online computational burden. To overcome this limitation, this paper proposes a novel sequential neural network (SNN) based FRT control strategy for GFM converters. Leveraging a long short-term memory architecture, the proposed SNN is trained offline via imitation learning to approximate the optimal nonlinear control law of an FS-MPC model. This approach fundamentally shifts the heavy optimization process offline, reducing online execution to deterministic, microsecond-level inference. An adaptive virtual resistance mechanism is integrated to dynamically constrain the voltage reference, enabling the SNN to strictly limit the converter fault current to a safe threshold of 1.5 per unit. Extensive time-domain simulations demonstrate that the SNN intrinsically manages unbalanced sequence components during severe asymmetrical faults without requiring complex sequence extraction filters. Comparative analysis reveals that the proposed data-driven strategy not only minimizes computational latency but also enhances grid-compliant power quality, achieving a current total harmonic distortion of less than 2%, thereby outperforming the FS-MPC baseline.