运用机器学习方法预测空气中臭氧浓度

臭氧(O_3)浓度变化与天然源、移动源和点源的排放量存在某些隐含的关联。根据臭氧浓度变化的特性,基于污染源在线排放数据、气象监测数据以及空气质量监测数据构造特征,运用机器学习方法进行逐小时臭氧浓度预测。该方法不仅充分利用了臭氧浓度变化时序数据,而且考虑了气象条件变化对污染物浓度变化的影响,最重要的是将点源排放氮氧化物这一臭氧生成的重要前体物纳入模型考虑。在金砖国家领导人厦门会晤前后(2017年8月31日至9月9日),运用该方法对厦门市溪东、洪文、鼓浪屿和湖里中学四个大气自动监测站的臭氧小时浓度平均值进行滚动预报,比较准确地模拟出臭氧浓度的日周期性变化,同时对峰值和低谷能够进行较为有效的捕捉和刻画。按照《环境空气质量标准》(GB3095—2012)臭氧日最大八小时浓度平均值进行评价,四个站点均取得了90%的预报等级准确率。

Using Machine Learning Method for Predicting the Concentration of Ozone in the Air

There are some implicit associations between changes ofozone concentration and emissions from natural sources, mobile sources and point sources. According to the characteristics of ozoneconcentration change, this paper used machine learning method to predict the hourly ozone concentration based on the online discharge data of pollutants, meteorological monitoring data and air quality monitoring data. This method not only made full use of time series data of ozone concentration change, but also took into account the influence of the change of meteorological conditions on the change of pollutant concentration. The most important point was to consider the point source nitrogen oxides emission, an important precursor of ozone generation, into the model. In the period before and after BRICS XiamenSummit (from August 31, 2017 to September 9th), this method was applied to ozone 1 hour average concentration rolling forecast of four automatic atmospheric monitoring stations including Xidong, Hongwen, Gulangyu and Huli Middle School in Xiamen, the diurnal periodic variation of ozone concentration was accurately simulated and the peak and trough were also effectively captured and characterized.According to the average of maximum 8 hours concentration in the ozone day of "Ambient Air Quality Standard" (GB3095-2012), all four sites achieved the accuracy of forecast grade of 90%.