Algerian Journal of Electrical Systems and Sustainability

Machine Learning Enhanced Virtual Synchronous Generator Control For Grid Forming Inverters In Low Strength Power Systems

Auteurs : Khan Rony Ahmed . Mekhilef Saad . Shah Noraisyah Mohamed . Mubin Marizan . Akter Md. Parvez .

Résumé

Low-strength, resistive grids challenge grid-forming inverter stability because fixed controller parameters cannot accommodate changing conditions. This study develops a machine-learning-enhanced virtual synchronous generator (ML-VSG) that schedules virtual inertia (J) and damping (D), and benchmarks it against conventional droop and fixed-parameter VSG control. A 5-15-15-2 feedforward artificial neural network was trained offline on 500 simulation-generated samples to map Δf, ΔP, ΔQ, ΔV, and SCR to J and D; its weights and biases were frozen for online inference. Unified MATLAB/Simulink tests covered SCR values of 1.5, 3, and 8 and X/R ratios of 1, 3, and 7 under active- and reactive-power reference steps from 2 to 4 MW and from 0 to 3 MVAr, respectively. All controllers achieved zero steady-state active-power error, but their transient responses differed. At the most demanding condition (SCR = 1.5, X/R = 1), the ML-VSG limited active- and reactive-power overshoots to approximately 6.3% and 1.7%, with corresponding settling times of 1.2 and 1.5 s; it also restricted peak frequency deviation to ±0.01 Hz, maximum RoCoF to 0.3 Hz/s, and frequency settling time to 0.6 s. The fixed-parameter VSG produced active- and reactive-power overshoots of 42.5% and 25.0%, a 2.6 Hz/s RoCoF, and active-power and frequency settling times exceeding 10 s. Droop control remained stable but reached 13.8% active-power overshoot and 1.8 Hz/s RoCoF. Across all tested cases, the ML-VSG consistently suppressed coupling-induced oscillations and accelerated recovery, demonstrating that offline-trained online parameter scheduling can improve GFMI robustness without online retraining.

Mots clés

Grid-forming inverter ; Machine learning control ; Virtual synchronous generator ; Frequency stability ; Weak grid