A Machine Learning Based Approach for Frequency Response Prediction in Low Inertia Power System

The uncertainty in securely operating the power system has increased with large-scale integration of inverter based resources (IBRs). IBR-driven displacement of conventional sources results in diminishing inherent frequency support capability of such systems as, unlike conventional synchronous generators, IBRs do not inherently support the grid frequency. Therefore, with increasing vulnerability, there is a need for an accurate frequency prediction model that could help a grid operator better plan system resources and securely operate the power system. Against this backdrop, this paper initially provides a critical insight into the frequency response of IBR-dominated systems to highlight the limitation of using a linear prediction model. Following the critical analysis, the paper proposes a data extraction framework and XGBoost algorithm-based regression model to predict RoCoF, frequency nadir, and quassi steady-state frequency of the IBR-dominated power system. The proposed approach has been implemented on a modified IEEE-39 bus system, and its comparative analysis with other state-of-art algorithms supports the superiority of the proposed method.