A dataset generation toolbox for dynamic security assessment: On the role of the security boundary

This paper introduces an efficient dataset generation toolbox, aiming to substantially improve data-driven dynamic security assessment methods (DSA). As conventional DSA methods are becoming intractable due to the increasing power system complexity, data-driven DSA methods are gaining significant interest. However, computationally efficient and high-quality dataset generation, which is the cornerstone for these methods’ performance, remains a major challenge. Despite efforts with generic or importance sampling techniques to focus on operating points near the system's security boundary, systematic methods for sampling in this region remain scarce. This paper presents two main contributions. First, it introduces an open-source dataset generation toolbox to efficiently generate balanced datasets along the security boundary of the system. Second, considering the lack of systematic evidence on the impact of sampling near the security boundary, this paper performs a comprehensive assessment of how accurately capturing the security boundary affects the performance of data-driven DSA methods. In this paper, we consider AC steady-state feasibility and small-signal stability. Our case studies on the PGLib-OPF 39-bus and 162-bus systems demonstrate the importance of (i) including boundary-adjacent operating points in training datasets, and (ii) maintaining a balanced distribution of secure and insecure points.