Prediction Of Air Quality Based On Grey Neural Network Model

Air quality prediction is the basis of air quality evaluation, management anddecision-making. In view of shortage and high fluctuation of the air quality data, twogrey neural network models were established based on understanding of the basictheory of grey system and the basic principles and algorithms of artificial neuralnetwork. These two models were applied to the air quality forecast field. The mainresearch of this article was to explore some new approach to improve the precisionand reliability of air quality.In the condition that natural factors, economic development and national industrypolicies did not have great change, we combined Grey Model with artificial neuralnetwork to establish an integration model. Then, improved grey neural network model（IGNNM） was presented by improving this combination model. This model can takeadvantage of GM（1,1） which need fewer data and BP neural network that havenonlinear fitting, and make up the deficiencies that GM（1,1） model has fluctuationspoor data fitting and BP neural network needs large sample to track data change. Asraw data, the annual average value of PM₁₀, SO₂and NO₂from2001to2008inTianjin city was used for modeling to simulate. Meanwhile, the concentration of PM₁₀,SO₂and NO₂in2009to2011was forecasted to check the precision of this model.Finally, the air quality of Tianjin between2012and2016was predicted by using thisproposed model. The results showed that in the cases of little change in air qualitydata, the model mentioned in this paper can be employed in air quality prediction as ithas less relative simulation error and higher reliability, compared with grey model andtraditional grey neural network model.In consideration of each factor that affects the air pollutants content, grey neuralnetwork combination model based on factor analysis （FA-GNNM） was established.FA-GNNM fully considers the influence of pollution source strength to pollutants.Modeling is more reliable. The annual average value of PM₁₀, SO₂and NO₂from2001to2008in Tianjin and the corresponding data of its influencing factors from2001to2010were used for modeling. The concentration of PM₁₀, SO₂and NO₂in2008to2010was forecasted to check the precision of this model. The relativeprediction error is less than5%, which indicates this prediction method is feasible. Comparing the two forecasting methods, the conclusions are as follows: IGNNMis easy to obtain data and modelize, but it don’t consider the influence of pollutionsource strength on pollutants. When air quality has large change, the predictionaccuracy would not be satisfying. FA-GNNM is more reliable because this modelfully considers the influence of pollution source strength. However，FA-GNNM datastatistical work is complex, meanwhile, data acquisition is difficult. In the futureresearch, these problems should be improved.