Non-Euclidean Grid-Partitioning to Mitigate Cascading Risk in Multi-Infeed HVDC System

The presence of multiple high voltage direct current (HVDC) systems in close proximity creates voltage-related cascading risks that are not adequately addressed by conventional grid-partitioning. This paper proposes an improved partitioning scheme to mitigate these new risks in addition to conventional objective of preventing parallel power flow from transferring adversely. Unlike classical partitioning approach, which relies solely on either optimization or clustering, our proposed bi-level architecture includes an additional HVDC clustering before optimization. However, this paper innovatively reveals that the distribution of correlation data to be used in clustering is non-Euclidean due to unusual equivalent reactance different from the normal operating condition, resulting from HVDC station's reactive power control. This non-Euclidean distribution makes heuristic clustering algorithms infeasible. To address this issue, an alternative solution is proposed to embed the correlation data into a dimension-reduced eigenspace spanned by selected eigenvectors, allowing clustering to be performed. The optimization implementing other objectives inherits the results of HVDC clustering as constraint, and the graphic betweenness weighted by power flows is presented to promote efficiency. Our proposed scheme is validated using cases studied in modified IEEE 118-bus benchmark system and practical regional grid, demonstrating its effectiveness in mitigating cascading risks in multi-infeed HVDC systems.