基于GA-ANN改进的空气质量预测模型

Improved GA_ANN Model for Air Quality Forecasting

基于人工神经网络的空气质量预测模型优于传统的逐步回归模型,但由于性能差异不明显而较少在空气质量预报中应用.设计了将遗传算法和神经网络算法相结合的基于GA-ANN的空气质量预测模型,并利用天津市2003—2007年气象和污染物监测资料对该模型进行验证.对2007年全年的ρ(SO2),ρ(NO2)和ρ(PM10)进行预测,预测值与实测值的相关系数分别为0.899 6,0.828 3和0.600 0.与一般的人工神经网络预测模型相比较,GA-ANN模型将空气质量等级预报的准确率从77.57%提高到79.67%.GA-ANN模型可结合其他方法进行日常空气质量预报.

The forecast precision using the artificial neural networks （ANN） approach is not obviously better than using a normal stepwise regression method, as indicated by a number of studies. A GA ANN Air Quality Forecasting Model was developed, which integrated the genetic algorithm （GA） and the neural network algorithm. Air quality monitoring data and meteorological data from 2003 to 2007 in Tianjin City were used in performance tests of the air quality forecasting. The results based on the data of 2007 indicated that the correlation coefficient between the forecasted value and its monitoring value for the pollutants SO2, NO2 and PM10 was 0. 8996,0. 8283 and 0. 6000, respectively. Compared with the normal ANN Model, the forecasting precision of the GAANN Model for air quality grade was increased from 77.57% to 79.67%. It is expected that the GAANN model will be put into daily air quality forecasting work in association with other approaches.