Enhancing Frequency Stability of Low-Inertia Grids with Novel Security Constrained Unit Commitment Approaches
Conventional synchronous generators are gradually being replaced by low-inertia inverter-based resources. Such transition introduces more complicated operation conditions. Insufficient system inertia is one of the challenges which would lead to dramatical change in rate of change of frequency (RoCoF) and further results in under frequency load shedding as well as tripping of generator protection devices. Inertia estimation can ensure the accountability and reliability of inertia response through implementation of frequency control ancillary services. Existing power system inertia estimation approaches do not take such heterogeneity into consideration, and the risk of estimation error introduced by uniform model-based approach is underestimated. Hence, strategies such as dynamic inertia estimation method and machine learning-assisted inertia estimation approaches are introduced. Frequency related constraints have been also imposed in the conventional security-constrained unit commitment (SCUC) model to keep the minimum amount of synchronous inertia online and secure system frequency stability. However, in a large system inertia distribution may vary significantly, some certain areas are more susceptible to frequency deviation following G-1 contingency, posing risks of load shedding and generation trip. To address these issues, two novel strategies are proposed in this thesis to ensure locational frequency stability: (1) A novel location based RoCoF constrained SCUC (LRC-SCUC) model that can capture the locational frequency response characteristics and counteract the im-pact of system oscillation, guaranteeing the RoCoF security following a G-1 disturbance. A multiple-measurement-window method is introduced in this work to constrain highest value during oscillation. Simulation results demonstrate the effectiveness of proposed LRC-SCUC model. The results also show that deploying virtual inertia techniques not only reduces the total cost, but also improves the system market efficiency. (2) A generic data-driven frequency and RoCoF predictor is first trained to predict maximal frequency deviation and the highest locational RoCoF simultaneously based on a high-fidelity simulation dataset. And the derived frequency related constraints are then incorporated into machine learning assisted SCUC to guarantee frequency stability following the worst generator outage case while ensuring operational efficiency. In addition, sparse computation and an active rectified linear unit (ReLU) linearization method are implemented to further improve the algorithm efficiency while retaining solution quality.