方位聚散因子及其在PM2.5浓度预报中的应用

污染物浓度预报是应对大气污染问题的重要手段.现有的模式类预报方法受限于排放源清单的准确性,而在污染物排放源短期少变的条件下,基于气象要素的统计类预报方法是一种更具实用性的方法.但现有统计类预报方法的计算模型输入量缺乏对气象要素累积效应的表征,以及对气象因素影响大气污染物聚散过程的表征,严重影响了预报的精度.为此,本文提出了一种着眼于改善计算模型输入量的统计类PM_(2.5)浓度预报方法.该方法采用方位聚散因子作为计算模型输入量,既可表征出PM_(2.5)累积与消散的过程,又考虑了气象要素在一定时段内的累积效应,为提高预报精度奠定了良好的基础.同时,通过BP神经网络训练,本方法在方位聚散因子与PM_(2.5)浓度值之间建立起关联模型,从而完成对PM_(2.5)浓度值的准确预报.

Azimuth convergence-diffusion factor and its application in PM2.5 concentration forecasting

Contaminant concentration forecast is an important means to deal with air pollution problems. Existing model-based forecasting methods are limited by the accuracy of source lists. Under the condition that the pollutant emission sources are stable during short terms, the statistical forecasting method based on meteorological elements is a more practical method. However, the input of the calculation model of the existing statistical forecasting method lacks the characterization of the cumulative effect of meteorological elements and the characterization of the atmospheric pollutants gathering process caused by meteorological factors, which affects the accuracy of the forecast seriously.To this end, a statistical PM_2.5 concentration forecasting method that focuses on improving the input of computational models is proposed in the paper. The method uses the azimuth convergence-diffusion factor as the input of the computational model, which not only represents the process of PM_2.5 accumulation and dissipation, but also considers the cumulative effect of meteorological elements in a certain period of time, which lays a good foundation for improving the prediction accuracy. Through the BP neural network training, the method establishes a correlation model between the azimuth convergence-diffusion factor and the PM_2.5 concentration value, thus completing the accurate prediction of the PM_2.5 concentration value.