基于聚类分析的集中式供水水质安全风险评价

Cluster analysis-based risk evaluation for water quality safety of centralized water supply

目的探讨适宜于集中式供水领域特点的风险评价方法。方法选择我国1 302座集中式供水设施,并计算其风险指数。采用两阶段聚类计算最优分类数。采用系统聚类和K-means聚类两种方法,并以最优分类数和标准推荐的三分类分别对研究对象的风险指数进行聚类分析。结果两阶段聚类显示最优分类数为4类;四分类系统聚类结果显示,97.8%的对象为2类;三分类系统聚类结果显示,98.5%的对象为1类;四分类的K-means聚类结果显示,14.0%、16.0%、30.6%和39.5%的对象被分别分为1~4类;三分类聚类结果显示,26.3%、34.2%和39.5%的对象被分别分为1~3类。结论集中式供水风险评价适宜分为分高、较高、较低、低四类;K-means聚类方法更适宜应用于该领域。

Objective To find risk evaluation method which is applicable to centralized water supplies. Methods Risk indices of each water supply of 1 302 centralized water supplied which were selected based on cluster sampling principle in five counties in China were calculated. Two-phase clustering algorithm was used to prioritize cluster number. Systematic clustering and K-means clustering algorithm were employed to analyze the risk indices base on the prioritized cluster number and cluster number recommended by national technical standard respectively. Results Two-phase clustering analysis result showed that four clusters was the prioritized cluster number for research objects; Systematic clustering analysis results of four clusters showed that 97.8% of the total was categorized in the second cluster, which presented most objects were evaluated as middle risk level; Systematic clustering analysis results of three clusters showed that 98.5% of the total was categorized in the first cluster, which presented most objects were evaluated as low risk level; K-means clustering analysis result of four clusters showed 14.0%, 16.0%, 30.6% and 39.5% of the total were categorized in the 1st, 2nd, 3rd and 4th cluster respectively; K-means clustering analysis result of three clusters showed 26.3%, 34.2% and 39.5% of the total were categorized in the 1st, 2nd and 3rd cluster respectively. Conclusion Four clusters is testified more applicable for risk evaluation of centralized water supply, which indicates high, relatively high, relatively low and low risk level respectively. K-means clustering analysis is regarded as the more applicable algorithm in this field based on the data characteristics.