33 / 2023-04-20 22:24:02

公共场所室内PM2.5浓度与吸烟行为预测模型研究/Research on the predictive model of indoor PM2.5 concentration and smoking behavior in public places

公共场所；细颗粒物；吸烟行为；随机森林模型

摘要：

[背景] 公共场所是吸烟行为高发场所之一，但目前尚缺乏实时、准确、智能监测识别吸烟行为的技术手段。因此，急需探索建立公共场所室内吸烟行为发生的预测模型，以提升场所内控烟监督执法效率。

[目的] 探索基于室内空气实时在线监测PM_2.5浓度的吸烟行为预测模型。

[方法] 以上海市浦东新区为例，于2022年10-11月间，采用方便抽样的方法选择6家服务类场所和4家办公类场所，每家场所选取1个监测点，对室内PM_2.5和尼古丁浓度进行连续7天监测。同时，获取同期上海市环境监测站点公共场所室外PM_2.5浓度数据。对室内外PM_2.5浓度比较采用Mann-Whitney U检验，室内PM_2.5和尼古丁浓度采用Spearman秩相关分析分析。此外，结合公共场所室内视频和图像采集，应用随机森林模型构建室内PM_2.5浓度吸烟行为预测模型。

[结果] 本次监测的服务类场所室内PM_2.5平均浓度（97.5±149.3 µg/m³）显著高于办公类场所（19.8±12.2µg/m³）(P<0.001)，且与同期室外大气PM_2.5浓度相比，服务类场所室内外PM_2.5浓度比值（I/O比值）在1.2~19.0之间，提示室内存在较强的PM_2.5污染源。进一步与尼古丁浓度对比，发现10家公共场所室内PM_2.5平均浓度排序与对应一周内尼古丁累积浓度排序一致（r_s=0.969，P＜0.001），其中室内PM_2.5与尼古丁浓度排名前三位均为：网吧、棋牌室、KTV。针对网吧场所开展为期7天的室内人群吸烟行为视频和图像采集，并与同期实时监测的PM_2.5进行联动效应分析，构建随机森林模型，发现同时刻PM_2.5对吸烟行为预测模型AUC为0.66，吸烟行为发生四分钟后的PM_2.5浓度预测模型AUC达0.72。

[结论] 公共场所室内PM_2.5浓度与吸烟行为高度关联，基于室内实时监测PM_2.5浓度初步构建了吸烟行为预测模型，具有较高的准确率，为公共场所控烟监督执法提供了新的途径。



关键词：公共场所；细颗粒物；吸烟行为；随机森林模型



Abstract:

[Background] Public places are one of the high-incidence areas for smoking behavior, but currently lack real-time, accurate, and intelligent technology for monitoring and identifying smoking behavior. Therefore, it is urgently needed to explore and establish a predictive model for indoor smoking behavior in public places to improve the efficiency of smoking control supervision and law enforcement.

[Objective] To explore a predictive model for smoking behavior based on real-time online monitoring of PM_2.5 concentration in indoor air in public places.

[Methods] Taking Pudong New Area in Shanghai as an example, six service places and four office places were selected using convenience sampling method from October to November 2022. One monitoring point was selected for each venue, and indoor PM_2.5 and nicotine concentrations were continuously monitored for seven days. At the same time, outdoor PM_2.5 concentration data for public places in Shanghai were obtained from the city's environmental monitoring stations during the same period. Mann-Whitney U test was used to compare indoor and outdoor PM_2.5 concentrations, and Spearman rank correlation analysis was used to analyze indoor PM_2.5 and nicotine concentrations. Additionally, a random forest model was applied to construct a predictive model for indoor smoking behavior based on indoor PM_2.5 concentration data and video and image acquisition in public places.

[Results] The average indoor PM_2.5 concentration in service places (97.5±149.3 µg/m³) was significantly higher than that in office places (19.8±12.2µg/m³) (P<0.001). Compared with the ambient PM_2.5 concentration outdoors, the indoor/outdoor PM_2.5 concentration ratio (I/O ratio) in service places ranged from 1.2 to 19.0, indicating a strong indoor PM_2.5 pollution source. Furthermore, when compared with nicotine concentrations, it was found that the ranking of indoor PM_2.5 concentration averages in 10 public places was consistent with the corresponding ranking of accumulated nicotine concentrations within a week (r_s=0.969, P＜0.001). Among them, the top three places ranked by indoor PM_2.5 and nicotine concentration were internet cafe, Chess and card room, and KTV. In the case of internet cafe, indoor crowd smoking behavior videos and images collection were carried out for 7 days, and linkage effect analysis was conducted with real-time monitoring of PM_2.5 during the same period. A random forest model was constructed, and it was found that the AUC of the PM_2.5-based smoking behavior prediction model was 0.66 at the same time, and the AUC of the PM_2.5 concentration prediction model four minutes after smoking behavior occurred was 0.72.

[Conclusion] The indoor PM_2.5 concentration in public places is highly correlated with smoking behavior. Based on real-time indoor PM_2.5 monitoring, a preliminary smoking behavior prediction model was constructed with high accuracy, providing a new approach for controlling smoking and law enforcement in public places.