Abstract:

In this paper, based on the PM_2.5, PM₁₀, SO₂, NO₂, CO, O₃ pollutant concentration data from 8 environmental monitoring stations of Chengdu and the T639 global medium-term numerical forecast model products from March of 2014 to February of 2017, the forecast model for the five of the above-mentioned six pollutant except O₃ in winter in Chengdu was built by using the recursive feature elimination (RFE) and the random forest method which are superior to the traditional statistical method, and its forecasting performance was assessed.The results show that the mean squared error (MSE) of the values of five pollutants forecasted by the RFE model are 47.58 μg·m^-3, 72.10 μg·m^-3, 8.87 μ·m^-3, 0.59 mg·m^-3, 19.84 μg·m^-3, and those by the random forest model are 23.94 μg·m^-3, 20.98 μg·m^-3, 2.40 μg·m^-3, 0.16 mg·m^-3, 8.09 μg·m^-3, which proves that the performance of the random forest model is better than that of the RFE model in the pollutant concentration forecast, indicating that the random forest method has a good performance and can provide the technical support for the air quality forecast business and the basis for the air pollution prevention and control in winter in Chengdu.

Key words: Air pollution forecast, Recursive feature elimination, Random forecast method