近三年四川省空气污染物浓度与天气因素联合分析

Joint Analysis of Air Pollutant Concentration and Weather Factors in Sichuan Province in the Past Three Years

摘要: 为了深入研究四川主要空气污染物浓度(PM_2.5, PM₁₀)与天气因素和其他污染物(CO, NO₂, O₃, SO₂)的相互影响，本文选用了2016~2018四川14个地区空气质量站点监测日数据集以及对应站点的中国气象台站地面气象资料日值数据集(V3.0)，运用相关分析的方法计算了主要空气污染物与气象要素(温度，气压，降水，相对湿度，风速)和其他污染物浓度的相关系数，并运用多元线性回归与BP神经网络的统计方法分别建立PM_2.5浓度预测模型，进而对比了两种模型的预报效果。结果表明：1) 主要污染物浓度与天气因素和其他污染物浓度的相关性在不同地区有着较大差异。总体来看，主要空气污染物浓度与温度、气压、降水有显著相关关系；2) 运用多元线性回归和BP神经网络两种方法分别建立的PM_2.5浓度预测模型显示，在相同数据条件下，BP神经网络预测的预报效果相较于多元线性回归更具优越性。

Abstract: In order to further study the interaction between the main air pollutant concentrations in Sichuan (PM_2.5, PM₁₀) and weather factors and other pollutants (CO, NO₂, O₃, SO₂), this paper selects the air quality monitoring daily data sets of 14 regional air quality stations in Sichuan from 2016 to 2018 and the daily value data sets (V3.0) of surface meteorological data of China Meteorological Obser-vatory of corresponding sites, using relevant analysis methods to find the main air pollutants corre-lation coefficients with meteorological elements (temperature, barometric pressure, precipitation, relative humidity, wind speed) and other pollutant concentrations, and using multiple linear re-gression and BP neural network methods to establish the PM_2.5 concentration prediction model, and then compare the prediction effects of the two models. The results show that 1) in the past three years, the correlation between the concentration of main pollutants and weather factors and the concentration of other pollutants varies greatly in different regions. Overall, the concentration of major air pollutants is significantly correlated with temperature, air pressure, and precipitation. 2) The PM_2.5 concentration prediction models established by multiple linear regression and BP neural network methods show that under the same data conditions, the prediction effect of BP neural network prediction is superior to multiple linear regression.