基于等效磁化曲线智能识别的变压器保护原理

Equivalent Magnetization Curve Intelligent Recognition Based Transformer Protection

提出一种基于等效磁化曲线智能识别的变压器保护原理。铁心的动态行为能够从本质上反应变压器的运行状态,磁化曲线的几何特征是变压器铁心动态行为的外在表现。首先,在研究磁化曲线几何特性的基础上,构建基于励磁支路电压-差动电流(U-I)的等效磁化曲线,并分析等效磁化曲线与变压器运行状态的对应关系;其次,构建以等效磁化曲线倾斜角度、椭圆率以及长轴数值为输入,以变压器运行状态为输出的BP神经网络模型,生成基于BP神经网络的变压器内部故障识别算法;最后,利用大量数字仿真和动模实验数据对所提算法进行验证,结果表明,保护方案从铁心动态行为出发,利用少量数据并结合传统的监督学习算法即可准确地判断变压器运行状态,仿真及动模实验数据的正确动作率均达到了100%。特别地,该方法具有良好的泛化能力,对CT饱和等场景具有良好的适应性,可以直接作为由铁磁材料构成的电力变压器的主保护,具有良好的应用前景。

This paper proposed an equivalent magnetization curve-based transformer protection. The transformer operation states are essentially affected by iron core and intuitively demonstrated by magnetization hysteresis loop. Firstly, based on the analysis of magnetization hysteresis loop, the correspondence between equivalent magnetization curves and operation states was shown in this paper. Secondly, several extracted geometric characteristics were used as input to BP neural network, which was trained with a small amount of training data to identify transformer operation states. Finally, digital simulation and dynamic-model experiments were conducted to verify the proposed scheme. The results of 100% showed that the classification model with a small amount of training data could accurately identify transformer operation states. Moreover, this classification model had higher generalization ability, not affected by CT saturation and over-excitation. The proposed scheme solves effectively the poor performance of AI technique in power system, which has a great application value.