常用大气污染预报模式的应用及比较分析

针对丹东市采暖期SO2污染的实际情况及气象因子的关系,建立了逐步回归、偏最小二乘回归、主成分回归和BP神经网络等4种常用的大气污染预报模式,并在实际预报中进行了模拟、试报和应用,结果发现,各个模式模拟值与实际值的变化趋势基本一致,BP神经网络方程和偏最小二乘回归方程的预报值与实际值的接近程度要好于逐步回归方程和主成分回归方程.

Four Common Air Pollution Forecast Modes: Application and Comparison

To forecast SO_2 pollution of during the heating period in Dandong city, four common air pollution forecast modes based on stepwise regression, partial least-square regression, principal component regression and BP neural network were established according to the relationship among weather factors. The results of simulation, test forecast and practical application show that the simulation values tend to be basically identical with the practical ones, in which the forecast values of modes of BP neural network and partial least-square regression are better.