Distributed Generation (DG), though offers many advantages in making modern power systems smart ones, unveils many technical problems in terms of its control and operation. Islanding is one such problem which represents the loss of grid condition during which DG continues to feed the load. Due to its harmful effects, islanding has to be differentially detected from other power system transients in minimum possible time. Since the islanding event immediately affects the reactive power level in the system, such a reflection in reactive power signal at Point of Common Coupling (PCC) is utilized to extract the significant information regarding its occurrence. Hence a novel islanding detection method based on the signatures extracted from reactive power using Empirical Mode Decomposition (EMD) is proposed. The effectiveness of the proposed method is tested in different power system models developed in MATLAB/SIMULINK and the simulation results demonstrate the efficiency of the method in islanding detection within two cycles time Also, the results of various test cases of load parameter combinations evidence that this method has very near zero Non-Detection Zone (NDZ) with no effect on power quality. Further, the proposed method is validated through extensive simulations on IEEE 14 bus and applicability on PV systems are tested using modified IEEE 14 bus model. By suitably setting the detection threshold, the proposed method can distinguish islanding event from the non-islanding ones. For making the system autonomous, K means clustering is combined with EMD for the detection and classification of islanding from other power system transients.

分布式发电（DG）尽管在使现代电力系统变得智能化方面具有许多优势，但在控制和操作方面却暴露出许多技术问题。孤岛就是一个这样的问题，它代表了电网条件的损失，在此期间DG继续给负载供电。由于其有害影响，必须在尽可能短的时间内与其他电力系统瞬变进行差异检测。由于孤岛事件立即影响系统中的无功功率水平，因此在公共耦合点（PCC）处对无功功率信号的这种反射可用于提取有关其发生的重要信息。因此，提出了一种新的基于经验模态分解（EMD）的基于从无功功率提取的特征的孤岛检测方法。在MATLAB / SIMULINK开发的不同电力系统模型中测试了该方法的有效性，仿真结果证明了该方法在两个周期时间内进行孤岛检测的效率。此外，负载参数组合的各种测试案例的结果证明了这一点。该方法的非检测区（NDZ）几乎为零，对电能质量没有影响。此外，通过在IEEE 14总线上的大量仿真对所提出的方法进行了验证，并使用改进的IEEE 14总线模型测试了在光伏系统上的适用性。通过适当设置检测阈值，该方法可以区分出孤岛事件和非孤岛事件。为了使系统具有自主性，将K均值聚类与EMD相结合，以检测和分类来自其他电力系统瞬变的孤岛。