基于深度学习和改进K-means聚类算法的电网无功电压快速分区研究

随着电网规模的不断扩大,对整个大电网进行统一的电压调控变得越发困难。提出一种基于深度学习和改进K-means聚类算法的电网无功电压快速分区方法。首先建立电耦合强度矩阵反映系统节点间的电气耦合关系的强弱。然后采用深度学习中的稀疏自编码器,通过训练实现对输入的高维矩阵进行特征提取和降维。最后基于改进的K-means聚类算法用以对降维后的特征序列进行聚类分析,通过检验电气模块度值来确定最终的分区。以电气模块度、无功储备校验两个评价指标对电网分区质量进行评估。对IEEE39节点和IEEE118节点系统进行仿真分析,验证了所提方法在保证连通性以及充足的无功储备的的基础上,具有较高的电气模块度。

Research on fast partition of reactive power and voltage based on deep learning and an improved K-means clustering algorithm

With the continuous expansion of the power grid, it has become more and more difficult to perform unified voltage regulation on the entire grid. This paper proposes a fast reactive power and voltage partition method based on deep learning and an improved K-means clustering algorithm. First, the electrical coupling strength matrix is established to reflect the strength of the electrical coupling relationship between the nodes of the system. Then the sparse autoencoder in deep learning is used to realize feature extraction and dimensionality reduction of the input high-dimensional matrix through training. Finally, the improved K-means clustering algorithm is used to perform cluster analysis on the feature sequence after dimensionality reduction, and the final partition is determined by checking the electrical modularity value. The quality of power grid divisions is evaluated with two evaluation indicators: electrical modularity and reactive power reserve verification. The simulation analysis of IEEE39 and IEEE118 bus systems verifies that the proposed method has high electrical modularity on the basis of ensuring connectivity and sufficient reactive power reserve. This work is supported by the National Key Research and Development Program of China (No. 2018YFB0905105).