基于改进卷积神经网络拓扑特征挖掘的配电网结构坚强性评估方法

Structure Strength Assessment Method of Distribution Network Based on Improved Convolution Neural Network and Network Topology Feature Mining

提出一种基于拓扑特征分析和深度卷积神经网络的配电网网架结构坚强性评估方法。将考虑分布式电源出力不确定性的配电网运行状态与配电网拓扑特性相结合,构建涵盖可靠性、运行裕度和结构鲁棒性3个配电网结构坚强性要点的拓扑指标集,提升评估指标刻画配电网实际状态的能力;通过多通道融合和多级池化改进卷积神经网络,解决了传统方法无法自主挖掘评估指标数据特征,以及难以直接分析不同维度的评价指标和同一指标不同尺寸的数据这两方面问题。通过对华东地区某中压配电网进行算例分析,说明所提出的评估方法的有效性和优越性。

A distribution network structure strength assessment method based on topology feature analysis and deep convolution neural network was proposed. Considering both uncertainties of distributed generation output in network operation and network topological features, a topology feature index set assessing network structure strength from the perspectives of reliability, security and structural robustness was built to raise the accuracy in describing distribution network operation states. The traditional convolution neural network was improved by combining multi-convolution channel merging and multi-scale pooling. The proposed method was capable of autonomously mining concealed topological characteristics and analyzing indices of different dimensions and index data of different scales, which were hardly achieved by traditional approaches. A case study of a medium-voltage distribution network in Eastern China was used to verify the validity and superiority of the proposed topology feature indices and the structure strength assessment model.