数据驱动的电力系统关键断面筛选

Data-driven Searching for Power System Key Transmission Sets

关键断面筛选对电力系统运行分析与控制至关重要。现代电力系统日益规模化和复杂化,传统的物理模型驱动的筛选方法计算效率偏低,难以满足在线分析需求,而现有的数据驱动算法难以兼顾模型性能、可解释性和发现新关键断面的能力。针对上述问题,该文提出了一种基于神经网络可解释性的数据驱动算法。首先,建立了基于多通道长短时记忆(long short-term memory,LSTM)神经网络的暂态稳定分析模型,根据故障后短时受扰曲线类型建立多个特征提取通道,实现功角稳定性预测。接着,采用基于积分梯度的神经网络可解释方法探究线路功率与暂态稳定性之间的关系,结合输电断面的拓扑约束和潮流流向约束,筛选出对暂态稳定影响较大的关键断面。最后,在IEEE39节点系统中对所提算法进行了测试,验证了算法的正确性,该算法兼顾了模型表达能力与可解释性,具备发现新关键断面的能力,计算效率高,可用于在线关键断面筛选。

The key transmission sets searching is essential to the real-time analysis and control of the power system.With the increasing scale and complexity of the power system,the low efficiency of the traditional physics-driven searching methods hardly meets the demands of the online monitoring,and the existing data-driven methods can neither meet the requirements of the model performance,the interpretability and the ability to discover the new key transmission sets simultaneously.To solve these problems,a data-driven algorithm based on the neural network interpretability is proposed in this paper.A multi-channel Long/Short-term Memory(LSTM)neural network is firstly established to realize the prediction of transient stability from different kinds of disturbed curves and global characteristic curves.Then,the integral gradient interpretable algorithm is used to explore the relationship between the input power and the transient stability.Combining with the topology constraints and the power flow constraints,the key transmission sets greatly influencing on the transient stability are screened.Finally,the validity of the method is verified by an example in the IEEE-39 system.This method maintains the expressiveness and interpretability of the model,and also has the ability to discover the new key sections,greatly improving the calculation efficiency and contributing to the online searching for the key transmission sets of the power system.