基于时序轨迹特征学习的暂态电压稳定评估

以先进机器学习方法等为代表的人工智能技术在增强现代电网安全稳定态势感知能力方面展现出巨大的潜力。针对在线暂态电压稳定评估的传统难题,提出基于时序轨迹特征学习的稳定评估方法。通过分析系统能量函数与暂态响应轨迹的相关性,给出学习过程输入数据选取的理论依据。在时序轨迹Shapelet变换基础上,提出以刻画系统稳定/失稳案例关键局部轨迹差异为核心的特征学习方法及稳定评估方案。双机四节点系统和南方电网中的算例测试结果表明,除了实现可靠的稳定监测和评估,还可充分利用文中方法的可解释性从数据层面剖析特定系统的失稳模式和规律。

Transient Voltage Stability Assessment Based on Sequential Trajectory Feature Learning

The boom of artificial intelligence, including advanced machine learning techniques, brings great potential into situational awareness enhancement for modern power grids. Aiming at tackling the conventional challenge of online transient voltage stability assessment, this paper develops a novel stability assessment approach based on sequential trajectory feature learning. By analyzing the correlation between system energy functions and transient trajectories, justification of sequential input data selection is firstly provided. With the help of shapelet transformation, a sequential trajectory feature learning method that captures representative local trajectories to discriminate stable cases from unstable ones is then proposed. Based on it, a stability assessment scheme is comprehensively constructed. Numerical test results on a two-machine four-bus system and China Southern Power Grid demonstrate the reliability of the proposed approach for online stability monitoring and assessment. Besides, its interpretability helps to anatomize specific system instability patterns and mechanism from the data-analytic perspective.