基于贝叶斯网络的集中空调系统冷却水中嗜肺军团菌生长繁殖风险预警模型研究

Early Warning Model for Risk of Legionella pneumophilia’s Growth and Reproduction in Cooling Water, Central Air Conditioning System Based on Bayesian Network

目的建立冷却水中嗜肺军团菌生长繁殖风险预警模型,实现对军团菌病预防控制的快速响应并为制定集中空调系统卫生管理政策提供科学依据。方法该研究主要包括现场监测及模型模拟两部分内容。在上海市选取使用集中空调的公共场所100家,在江苏省南京、苏州、常州三城市选取使用集中空调的公共场所148家,筛选冷却水及相关环境指标进行现场监测。通过上海市现场所测各指标数值辅以文献报道先验知识及专家经验建立贝叶斯网络模型,以江苏省3座城市的监测数据对模型预警结果进行验证。结果上海市冷却水嗜肺军团菌阳性率达到了79%,江苏省三城市阳性率达到了75%,基于贝叶斯网络的预警模型显示,从定性关系上看,对嗜肺军团菌最具影响的因素是日照强度及浊度,其中浊度又分别与余氯、溶解性总固体、菌落总数、电导率、水温等因素存在着联系。在定量层面上,可预测出嗜肺军团菌阳性或阴性概率水平,按照预测阳性概率>0.8,0.5～0.8,<0.5分为高、中、低3个预警等级。模型评价结果显示,该模型对冷却水中嗜肺军团菌的预测敏感性水平达到85.2%(52/61),特异性水平为8.3%(1/12)。结论本模型可以满足目前嗜肺军团菌预警的要求,阴性结果预测准确度有待提高,可通过加大监测样本量,特别是阴性结果样本量予以改进。

Objective To establish an effective early warning model for the risk of Legionella pneumophilia＇s growth and reproduction in cooling water for the quick response for LegioneUa＇ disease control and prevention, and the significant scientific basis for developing the health management policy for central air conditioning systems. Methods Field monitoring and modeling two parts of work were included in this study. One hundred public places in shanghai and 148 places in Jiangsu province were selected to be the field work monitoring site. The monitoring items included cooling water items and related environmental items. Next, the bayesian network model was set up based on the field monitoring results in Shanghai and the knowledge reported from the literature. The field monitoring results were used to validate the modeling results. Results The positive rate of Legionella pneumophilia in cooling tower attained 79% and 75% in shanghai and Jiangsu respectively. From the qualitative perspective, the most influential factors were the sunlight intensity and the turbidity. The turbidity was also related to residual chlorine, total dissolved solid, total bacterial count, conductivity and water temperature. It provided an instructive thought for controlling the growth and reproduction of Legionella pneumophilia effectively. From the quantitative perspective, the modeling calculation provided the positive or negative probability level of Legionella pneumophilia by inputting the relevant parameters＇ values. According to the probability distribution, 3 early warning levels were affirmed. The model evaluation results showed that the sensitivity and specificity of forecasting Legionella pneumophilia were 85.2% and 8.3% respectively. Conclusion The model employed in the present paper can meet the early warning requirements currently. The negative forecasting accuracy should be improved by increasing monitoring sample size, especially the sample size of negative results.