Regulation of Disturbance Magnitude for Locational Frequency Stability Using Machine Learning

Power systems must maintain the frequency within acceptable limits when subjected to a disturbance. To ensure this, the most significant credible disturbance in the system is normally used as a benchmark to allocate the Primary Frequency Response (PFR) resources. However, the overall reduction of system inertia due to increased integration of Converter Interfaced Generation (CIG) implies that systems with high penetration of CIG require more frequency control services –which are either costly or unavailable. In extreme cases of cost and scarcity, regulating the most significant disturbance magnitude can offer an efficient solution to this problem. This paper proposes a Machine Learning (ML) based technique to regulate the disturbance magnitude of the power system to comply with the frequency stability requirements i.e., Rate of Change of Frequency (RoCoF) and frequency nadir. Unlike traditional approaches which limit the disturbance magnitude by using the Centre Of Inertia (COI) because the locational frequency responses of the network are analytically hard to derive, the proposed method is able to capture such complexities using data-driven techniques. The method does not rely on the computationally intensive RMS-Time Domain Simulations (TDS), once trained offline. Consequently, by considering the locational frequency dynamics of the system, operators can identify operating conditions (OC) that fulfil frequency requirements at every monitored bus in the network, without the allocation of additional frequency control services such as inertia. The effectiveness of the proposed method is demonstrated on the modified IEEE 39 Bus network.