Any disturbance in the grid affects the real and reactive power outputs of the generators. The capability of producing real and reactive power is usually known as capability curve. The conventional static capability uses predefined operating constraints pertaining to mechanical power, rotor angle, terminal voltage, rated power factor and field voltage of the generator. However, the generators connected to the grid have dynamic operating states which can be predicted in real-time using data from phasor measurement unit (PMU). The generator transient stability is important from power system stability point of view because tripping of generators due to instability creates cascading effects. The present work proposes the real-time monitoring of dynamic state of generators through dynamic capability curve using rotor angle. The rotor angles are not directly measured by the PMUs and hence are being estimated using extended Kalman filter. Values of rotor angle are used for principal component analysis (PCA) for identifying coherent and non-coherent generators. Any non-coherent generator beyond the dynamic capability curve limits is a clear indication of eventually becoming an unstable generator and has been termed as most critical generator. The proposed scheme employs multiple artificial neural network to incorporate inference from dynamic capability curve as well as that from PCA for identifying critical generators as well as predicting the degree of criticality of marginally critical generators within a time window of 12 cycles after occurrence of fault. The simulation results using case studies from IEEE-39 bus system validates the efficacy of the proposed methodology.

电网中的任何干扰都会影响发电机的有功和无功输出。产生有功和无功功率的能力通常称为能力曲线。传统的静态能力使用与机械功率，转子角，端子电压，额定功率因数和发电机的励磁电压有关的预定义操作约束。但是，连接到电网的发电机具有动态运行状态，可以使用相量测量单元（PMU）的数据进行实时预测。从电力系统稳定性的角度来看，发电机的暂态稳定性很重要，因为由于不稳定性而导致的发电机跳闸会产生级联效应。本工作提出了利用转子角度通过动态能力曲线对发电机动态状态进行实时监测的方法。转子角不是由PMU直接测量的，因此使用扩展的卡尔曼滤波器进行估算。转子角的值用于主成分分析（PCA），以识别相干发电机和非相干发电机。任何超出动态能力曲线极限的非相干发电机都清楚地表明最终会变成不稳定的发电机，并被称为最关键的发电机。提出的方案采用了多个人工神经网络，结合了动态能力曲线和PCA的推论，以识别关键发电机，并预测故障发生后12个周期内边际关键发电机的临界程度。