改进的给水管网节点K均值空间聚类

Improved K average spatial clustering method for nodes of water distribution system

利用自适应精英保留遗传算法优选初始聚类中心,信息熵确定属性权重改进K均值空间聚类,应用于给水管网节点聚类.实例验证表明,改进的K均值空间聚类方法在聚类精度、稳定性、耗时、权重计算方面具有明显的优越性:相对于传统的K均值空间聚类,自适应精英保留策略遗传算法和信息熵确定权重的K均值空间聚类得到的类内距离均值和标准差分别由6.92\1.06下降至4.39\0,聚类精度和稳定性均有较大程度提高;普通遗传算法、自适应精英保留策略遗传算法和模拟退火遗传算法优化的K均值空间聚类消耗时间分别为342、123、383s,自适应精英保留策略遗传算法消耗时间最短;管网拓扑图表明,信息熵权重能客观计算属性权重,结果更加合理.

The adaptive elitist genetic algorithm was introducted to optimize the choice of the initial cluster centers. The informationg entropy was combined to objectively determine the attributes＇ weights, which can improve K-average spatial clustering for nodes of water distribution system. The case study prove that adaptive elitist genetic algorithm K average spatial clustering has obvious advantages in clustering accura- cy, stability, elapsed time and weight choice. Relative to traditional K-average spatial clustering, the aver- age value and standard deviation of inner-class distance resulting from adaptive elitist genetic algorithm K average spatial clustering respectively decrease from 6.92／1.06 to 4.39 ／0. The clustering accuracy and stability can be apparently improved. Elapsed time of genetic algorithm K average spatial clustering, adap- tive elitist genetic algorithm K average spatial clustering and simulated annealing genetic algorithm K aver- age spatial clustering were respectively 342,123,383 s, Elapsed time of adaptive elitist genetic algorithm K average spatial clustering is the shortest among the three methods ; Network topology shows that the infor- mation entropy can objectively determine attributes;weights, and the result is more reasonable.