Machine Learning Based Inertia Constrained Unit Commitment for Frequency Stability of Renewable Rich Power System

The power system has been facing significant challenges in maintaining system security with the increasing uptake of renewable energy (RE). Of various issues arising from the growing share of inverter-based resources (IBRs), frequency instability is one of the major concerns in low-inertia power systems. The high share of the IBRs reduces the system inertia by displacing the conventional generators, and their lack of participation further compromises the frequency stability. Amid the ongoing advancement of controllers to tackle declining system inertia, the day-ahead operational planning for ensuring the necessary system inertia is about to play a pivotal role in secure grid planning. Modeling the required inertia in day-ahead unit commitment (UC) is challenging as system inertia requirements are dynamic and change with system conditions. Against some of the existing approaches to model inertia in UC, a mixed integer programming-based Decision Tree (DT) model has been developed to co-optimize the inertia requirement in UC. The DT regressor has been trained on historical data of the modified IEEE39 system, which has been incorporated into the UC framework for rate of change of frequency (RoCoF)-informed resource planning. The performance of the proposed model has been compared with existing techniques, and the analysis suggests that the proposed machine learning-based model performs better than the available model in terms of better emulation of system behavior and reduction in overall operational cost.