Scholarlink

Operational patterns have become increasingly complex with the large-scale integration of renewable energy and the continuous expansion of power grid systems. Therefore, stability assessments based on historically constructed security boundaries no longer meet current needs. This paper introduces an adaptive probability prediction model for security boundary violations based on incremental learning. Initially, unsupervised clustering is employed to identify how different line load states affect boundary fitting. Subsequently, a boundary set is constructed, and a probability assessment model is developed in batches using incremental learning to identify potential boundaries that may exceed limits in real-time conditions. Finally, confidence score distributions are analyzed to filter samples that meet evaluation criteria, which are added to the training pool for model adaptation. The model demonstrates robust adaptability, enabling rapid assessments and responses to online data, and its effectiveness and accuracy are validated through tests on the IEEE 39-bus system and China’s practical power system.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Incremental Learning-Based Adaptive Probability Prediction Model for Security Boundary Exceedance

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yuan Zeng,
  • Wenya Xue,
  • Junzhi Ren,
  • Wei Xu,
  • Weilun Ni,
  • Pengfei Song,
  • Pengpeng Kang,
  • Wenhui Zhai

摘要

Operational patterns have become increasingly complex with the large-scale integration of renewable energy and the continuous expansion of power grid systems. Therefore, stability assessments based on historically constructed security boundaries no longer meet current needs. This paper introduces an adaptive probability prediction model for security boundary violations based on incremental learning. Initially, unsupervised clustering is employed to identify how different line load states affect boundary fitting. Subsequently, a boundary set is constructed, and a probability assessment model is developed in batches using incremental learning to identify potential boundaries that may exceed limits in real-time conditions. Finally, confidence score distributions are analyzed to filter samples that meet evaluation criteria, which are added to the training pool for model adaptation. The model demonstrates robust adaptability, enabling rapid assessments and responses to online data, and its effectiveness and accuracy are validated through tests on the IEEE 39-bus system and China’s practical power system.