A Novel Clustering Method for Extracting Representative Photovoltaic Scenarios Considering Power, Energy, and Variability

Abstract

Due to the significant uncertainty in photovoltaic (PV) power generation, grid operation scenarios with a high proportion of PV integration are complex and varied. To accurately extract representative scenarios for PV power generation, this paper proposes a novel clustering model that simultaneously considers PV power, energy, and variability. Compared to traditional clustering models that rely on Euclidean distance, the proposed clustering model not only takes into account the Euclidean distance, but also incorporates the daily PV power generation and the characteristics of PV power curves, enabling a more accurate quantification and analysis of the impact of PV on the electricity networks. To solve the proposed clustering model, an alternating optimization algorithm is proposed, based on linear optimization, Lagrange multipliers, and eigenvalue decomposition. The highlights of this paper are the dual verification of the proposed method through theoretical proof and simulation examples. Theoretically, the computational complexity of the algorithm is illustrated, and the convergence of the algorithm is demonstrated. The proposed method is tested using real PV data from Australia and the IEEE 69-bus system, successfully generating 13 representative PV generation scenarios with a maximum similarity distance of the morphological trend as low as 0.3062, ensuring the most representative PV generation peak times.