高级搜索

长江中游PM2.5和O3复合污染气象特征分析及浓度预测模型研究

Study on the characteristics of PM2.5 and O3 compound air pollution and concentration prediction model in the middle reaches of the Yangtze river

    摘要
  • 摘要: PM2.5和O3是影响我国城市和区域空气质量的主要因子，探究其污染特征并对其浓度进行预测是大气复合污染防治的基础工作。首先，利用2015—2023年长江中游宜昌和武汉两个主要城市国控站PM2.5和O3质量浓度‌监测数据，分析PM2.5和O3复合污染特征；然后，利用机器学习模型可解释工具(shapely additive explanation，SHAP)，揭示气温、相对湿度、降水量、日照、风速等气象因子对PM2.5和O3浓度的影响及贡献；再构建基于融合门控循环单元(gated recurrent unit，GRU)等9种深度学习方法的PM2.5和O3浓度预测模型，并进行效果检验。结果表明：(1) 2015—2023年长江中游宜昌和武汉O3浓度每年平均依次升高3.89 µg·m−3和2.73 µg·m−3，夏、秋季升高更明显；PM2.5浓度则呈显著下降趋势，趋势率分别为−3.59 µg·m−3·a−1和−3.36 µg·m−3·a−1，冬、春季下降更显著，表明近年来PM2.5污染治理起到显著成效。(2) PM2.5和O3浓度月际间分别呈“U”和“M”型分布，两者呈弱的负相关关系。2015—2023年宜昌和武汉PM2.5和O3浓度“双高”天数分别为60 d和39 d，主要集中在2—5月和10—12月之间，且呈年下降趋势(2023年略有升高)。(3) 对比分析9种深度学习预测模型表明，门控循环单元(gated recurrent unit，GRU)、双向门控循环单元(bidirectional gated recurrent units，BIGRU)、基于注意力机制门控循环单元(attention gated recurrent unit，Attention-GRU)及基于注意力机制双向门控循环单元(attention bidirectional gated recurrent units， Attention-BiGRU) 共4种模型在宜昌和武汉PM2.5和O3浓度预测中效果较好，其中GRU运行时间最短，可有效提高PM2.5和O3浓度预测和服务的及时性。(4) 构建的基于融合深度学习回归预测模型与GRU相比，宜昌和武汉O3浓度预测均方根误差(root mean square error，RMSE)分别减小5%和8%，PM2.5浓度预测RMSE分别减小20%和16%。该模型对PM2.5和O3复合污染日预测Ts评分宜昌为60.00%，武汉为69.23%，可为长江中游宜昌和武汉两城市受气象条件影响的大气PM2.5和O3浓度预测及复合污染防治提供科学依据。

     

    Abstract: PM2.5 and O3 were the main factors affecting urban and regional air quality in China, exploring their pollution characteristics and predicting their concentrations were the basic work for the prevention and control of composite atmospheric pollution. The PM2.5 and O3 compound air pollution characteristics were analyzed using the concentration data of PM2.5 and O3 in two major cities, Yichang and Wuhan, in the middle reaches of the Yangtze River from 2015 to 2023. The machine learning model explainable tool was used to reveal the contribution of meteorological factors such as temperature, humidity, precipitation, sunshine, wind speed and other meteorological factors to the influence of PM2.5 and O3 concentrations．Based on nine deep learning methods including GRU, a regression prediction model was constructed for PM2.5 and O3 concentrations and the results were as follows.(1)The annual characteristics of O3 concentrations in Yichang and Wuhan was increased by 3.89µg·m−3 and 2.73 µg·m−3 on each year from 2015 to 2023, and the increase trend was more greater in summer and autumn; while the PM2.5 concentrations showed a significant decrease trend, with the trend rates of −3.59 µg·m−3·a−1 and −3.36 µg·m−3.a-1 respectively.The decrease was more notable in winter an spring, while indicated that the treatment of pollutant had achieved significant effect in recent years.(2)The monthly changes of PM2.5 and O3 concentrations was showed U and M features, and with a weak negative correlation between PM2.5 and O3. The compound air pollution days in Yichang and Wuhan were 60d and 39d from 2015 to 2023 respectively, mainly concentrated in February-May and October-December. The days was showed an decreasing trend annually, but higher in 2023 slightly.(3)A total of GRU, BIGRU, Attention-GRU and Attention-BiGRU models were performed better in predicting PM2.5 and O3 concentrations in Yichang and Wuhan, which GRU model had the shortest running time. The GRU model would improve PM2.5 and O3 concentration preciditon effectively.(4)Compared with GRU, the root mean square error for O3 concentration prediction of the regression prediction model was reduced by 5% and 8% in Yichang and Wuhan, meanwhile PM2.5 concentration prediction was improved by 20% and 16% respectively. Ts scores of PM2.5 and O3 was 60.00% in Yichang and 69.23% in Wuhan. The regression prediction model can provide scientific basis for the prediction of PM2.5 and O3 concentration and the control of compound pollution in Yichang and Wuhan cities which were affected by meteorological conditions.

     

    • HTML全文
    • 参考文献(39)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回