Skip to Content
ElectronicsElectronics
  • Article
  • Open Access

7 August 2026

19 Pages

Data-Driven Comprehensive Security Region Assessment for Hydro–Wind–Solar Hybrid Systems with HVDC Integration

,
,
,
,
and
1
Institute of Science and Technology, China Three Gorges Corporation, Beijing 100038, China
2
Laboratory of Hydro-Wind-Solar Multi-Energy Control Coordination, Wuhan 430014, China
3
School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, China
4
China Yangtze Power Co., Ltd., Wuhan 430014, China

Abstract

The massive integration of “hydro–wind–solar hybrid” generation transmitted via AC/DC hybrid grids with High-Voltage Direct Current introduces unprecedented challenges to power system transient stability. Traditional Dynamic Security Assessment heavily relies on time-domain simulation and high-density Monte Carlo sampling, which suffer from massive computational burdens and are strictly prohibitive for intra-day operational dispatch. To address this long-standing bottleneck, this paper proposes a novel data-driven comprehensive security region assessment framework based on an Active Learning query strategy and a Support Vector Machine surrogate model. By seamlessly coupling Python with the DIgSILENT PowerFactory simulator, the proposed AL algorithm actively queries and evaluates only the critical operating points near the stability margin, effectively avoiding redundant simulations in obviously safe or unsafe zones. Extensive simulations under a severe N-1-1 contingency demonstrate that the proposed framework can accurately map the non-linear boundaries of both transient rotor angle and short-term voltage stability constraints. Furthermore, the framework’s scalability is rigorously validated in complex 3D high-dimensional operational spaces. By integrating an ϵ-greedy exploration strategy with a Label Flip Rate (LFR) early stopping criterion, the proposed algorithm successfully overcomes the curse of dimensionality. Quantitatively, the proposed method achieves nearly identical boundary resolution using merely 65 physical simulations, as opposed to the 40,000 evaluations required by conventional high-density grid scanning. The total computational time is drastically reduced from 28.6 h to approximately 2.8 min, yielding a remarkable acceleration of over 600 times, making it highly suitable for near-online dynamic security monitoring.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.