Impact of Virtual Synchronous Generators on Frequency-Constrained Unit Commitment: A Neural Network-Assisted Framework for Assessing Rate of Change of Frequency

The increasing penetration of inverter-based resources (IBRs) into electric power systems and the retirement of synchronous generators (SGs) both reduce system inertia and challenge frequency stability. This is particularly evident in terms of the rate of change of frequency (ROCOF) measured over short, practically measurable time windows (P-ROCOF). Conventional frequency-constrained unit commitment (FCUC) uses a theoretically idealized ROCOF (T-ROCOF) constraint that underestimates the effect of the fast frequency response (FFR) provided by virtual synchronous generators (VSGs) on P-ROCOF, leading to over-provisioning of generator capacity and higher operating costs. This paper presents a neural network-assisted FCUC (NN-UC) framework that formulates a mixed-integer ROCOF constraint based on an NN trained to accurately predict P-ROCOF, and introduces a post-solution compensation procedure to correct NN prediction errors. We conducted one-year scenario studies on modified IEEE 9-bus and Japanese EAST 10-machine systems. Results show that NN-UC reduces both total operating cost and committed SG capacity compared to the conventional UC, which uses the T-ROCOF constraint. Even after worst-case compensation for NN’s ROCOF prediction errors, NN-UC still achieves net SG capacity reductions in most scenarios. These findings demonstrate that explicitly modeling the VSG’s FFR in FCUC yields both economic and operational benefits, while the compensation procedure ensures reliable operation.