基于广域时序数据挖掘策略的暂态电压稳定评估

Wide-Area Time Series Data Mining Based Transient Voltage Stability Assessment

在能源互联网背景下,大数据分析方法可为电力系统中一些传统难题提供新的解决思路。在基于大数据的暂态电压稳定评估中,针对动态变化趋势和特征难以准确捕获的问题,引入时间序列shapelet方法,从故障后PMU量测得到的动态序列中进行可靠的时序特征提取。通过融入错分代价的决策树算法,调整稳定/失稳样本的权重,使评估模型尽可能降低对失稳案例漏判的概率。Nordic系统算例对整体评估方案的测试表明,分类评估模型在保证高分类性能的同时,还可提供良好的可解释性,为特定系统失稳规律认知和在线监控提供进一步指导。

In context of Energy Internet, big data analytics provides possibility of tackling an array of traditional troublesome problems in power systems. In terms of big data-based transient voltage stability assessment, precise capture of critical evolution trends or states is a great challenge. To cope with it, a time series shapelet methodology, reliably extracting sequence features from dynamic post-contingency PMU measurements, is introduced. Additionally, by setting distinct weights for stable/unstable cases, misclassifying costs of classification learning are incorporated into decision tree algorithm, so that the assessment model would avoid mistaking unstable cases as far as possible. Test results on Nordic system for whole assessment scheme demonstrate that the assessment model not only holds a high performance of classification, but also offers considerable insights into the results, providing guidelines for further comprehension of specific system instability patterns and online monitoring.