供水管网计算分区方法的比较分析

Comparative analysis of calculation methods for water distribution system partitioning

摘要:

针对现有的供水管网计算分区方法研究多基于单个管网案例进行验证,缺乏不同案例、不同需求下的比较和适用性分析的问题,基于深度优先搜索-部分接近度算法（DFS-PCC）、快速迭代模块度的贪心算法（CNM）和遗传优化的谱聚类算法（GA-SC）3种方法,采用标准化互信息（NMI）、模块度、节点数均衡性、联络管数量等分区结果评价指标,在5个标准管网案例上比较其分区效果.比较分析时考虑了不同管网案例的拓扑结构、水源数量和类型、控制元件等固有属性差异,并讨论方法的赋权和分区数的确定问题.结果表明:DFS-PCC对区域供水特征明显、树状特性较高的案例分区,所得结果模块度较高,但联络管数量较多；CNM对区域供水特征明显、树状特性较高的案例分区,所得结果NMI较高,联络管数量较少；GA-SC在5个不同案例中,模块度较高、区域间节点数较均衡、联络管数量较少,总体适用性良好；使用该方法1/q的赋权形式,可有效选择流量和流量波动性大的管道作为联络管.

Abstract:

The existing research on the calculating methods for WDS partitioning is mainly based on single network cases, while it lacks comparison and applicability analysis for different network cases and various requirements. This paper compares the partitioning effects of depth-first search combined with partial closeness centrality algorithm (DFS-PCC), fast iterative modularity greedy algorithm (CNM), and spectral clustering optimized by genetics algorithm (GA-SC) in five benchmark cases. The comparative analysis was achieved by developing evaluation indicators such as normalized mutual information (NMI), modularity, the balance of nodes quantity, and the number of feed lines. The influences of network intrinsic properties including topology structures of the cases, the number and types of water sources, and control elements were considered. In addition, the selection of weights and the determination of partition numbers were studied. Results show that for the cases with obvious regional water supply and high tree-like characteristics, DFS-PCC had high modularity and a large number of feed lines, while CNM had high NMI and a few feed lines. In the five cases, GA-SC had high modularity, a balanced number of nodes, and a few feed lines, indicating better applicability. By utilizing the weights of 1/q, the pipes with large flow and flow fluctuation could be effectively selected as feed lines.