为了提高发电机失磁保护的选择性和速动性，提出了一种基于支持向量机（SVM）进行轨迹智能识别的发电机失磁保护方法。机端测量阻抗轨迹蕴含大量发电机运行信息，其运动特征能反映发电机的运行状态。首先对机端测量阻抗轨迹进行运动特征提取，将提取的运动特征序列分别进行统计学参数计算，形成24维特征；其次通过相关系数分析和前向序惯法进行特征选择，形成相应的训练输入特征空间，并采用遗传模拟退火算法对SVM进行参数寻优；最后通过仿真样本验证了该方法可准确识别失磁故障。相比传统失磁保护，该方法提高了发电机失磁保护动作的选择性和速动性。

In order to improve the selectivity and quick action of generator loss of field (LOF) protection, a method for LOF protection based on support vector machine (SVM) for intelligent identification of trajectory is proposed. The measured impedance trajectory of the generator contains lots of generator operation information, and the motion characteristics can reflect the generator's operating state. First, the global and local features of the measured impedance trajectory of the generator are extracted, and the extracted motion feature sequence is calculated separately for statistical parameters to form 24-dimensional features. Then through principal component analysis of the feature space to reduce the dimensionality, the corresponding training input feature space is formed, and genetic simulated annealing algorithm is used for the SVM to optimize the parameters support. Finally, sample simulation verifies that this method can accurately identify the LOF. Compared with the traditional LOF protection, the proposed method improves the selectivity and quickness of the protection.

国家重点研发计划项目(2016YFB0900503)