As a typical cyber-physical system, dispersed Phasor Measurement Units (PMUs) are networked together with advanced communication infrastructures to record the Distribution Synchrophasor (DS) which represents the states and dynamics of distribution power networks. Source information of DS is critical for many DS-based applications which is potentially vulnerable to data integrity attacks. To ensure the reliability of DS-based applications, it is imperative to efficiently authenticate the DS source locations before any DS data analytics is initiated. This letter presents a cost-effective method for accurate source identification by realising the multifractality of DS data. First, Multiscale Adaptive Multifractal Detrended Fluctuation Analysis (MSA-MFDFA) is executed to reveal the scale which possesses the most significant multifractality of the time-series DS. Subsequently, Adaptive Multifractal Interpolation (AMFI) is proposed to generate quasi high-resolution DS where unique time-frequency signatures are extracted. Such signatures are further fed into a deep learning model - deep forest for source identification. Experimental results using real-life DS of a distribution network illustrate the excellent performance of the proposed approach.

中文翻译：

---

基于多尺度自适应多分形去趋势波动分析的同步相量数据源识别

作为典型的信息物理系统，分散相量测量单元 (PMU) 与先进的通信基础设施联网，以记录代表配电网络状态和动态的配电同步相量 (DS)。DS 的源信息对于许多可能容易受到数据完整性攻击的基于 DS 的应用程序来说至关重要。为了确保基于 DS 的应用程序的可靠性，必须在启动任何 DS 数据分析之前有效地验证 DS 源位置。这封信提出了一种通过实现 DS 数据的多重分形来准确识别源的经济有效的方法。第一的，执行多尺度自适应多重分形去趋势波动分析（MSA-MFDFA）以揭示具有时间序列 DS 的最显着多重分形的尺度。随后，提出了自适应多重分形插值 (AMFI) 来生成准高分辨率 DS，其中提取了独特的时频特征。这样的签名被进一步输入深度学习模型——用于源识别的深度森林。使用配电网络的真实 DS 的实验结果说明了所提出方法的优异性能。