A Neural Network Approach to Adaptive Inference of Frequency Droop Curves in Power Systems with Solar PV Plants

Abstract

The provision of frequency droop (FD) curves is imperative in supporting communication-free grid operations, distinctly for power systems with utility-scale solar photovoltaic (PV) power plants. Besides, system compliant frequency/voltage excursions at PV power plant point of interconnection (POI) exert a high level of inherently stochastic dependence and parameter uncertainty, partially due to the PV power plant site-specific spatio-temporal meteorological conditions. Hence, a site-dependent variable FD has been modeled through controlled lab environment simulations with a real-time digital simulator, i.e., RTDS, and derived from integrating site-specific operating conditions imposed by PV-site steady-state loading levels during different hours of the day. The adaptive FD curve replicates close to linear FD droop deviations curve at each stable (reference) operating condition. For a more accurate characterization, this paper incorporates an additional layer of artificial intelligence, in recognition of the non-linearity of PV plant adaptive frequency droop curves. The results reflect on the effectiveness of a supervised machine learning approach, i.e., single neural networks (SNN) in the realization of the nonlinear adaptive frequency droop curve from the utility-scale solar PV power plant operation in a power system.