基于BP神经网络的贵阳市空气质量指数预报模型

采用贵阳市2013年1月1日—2015年12月31日的空气质量指数(AQI)日均值,常规的地面和高空观测资料,基于不同季节,调整BP神经网络的隐藏层个数和隐藏层节点数,建立不同的BP神经网络预报模型,进行参数检验,最终选取预报效果最好的模型带入实况进行检验。结果表明,夏季的预报效果最好,采用的模型TS评分为81.6%,平均绝对误差为9.1,正确率为97.4%,用该模型检验预报效果,实况和预报的相关系数为0.71,平均误差为9;而冬季的预报效果明显低于其他季节,采用的模型TS评分为65.7%,平均绝对误差为19.5,正确率为72.9%,用该模型检验预报效果,实况和预报的相关系数为0.79,平均误差为19。而且BP神经网络模型的预报效果同隐藏层个数与隐藏层节点数没有显著关系。

Prediction Model of Air Quality Index in Guiyang City Based on BP Neural Network

Different BP neural network forecast models for the AQI(air quality index) have been developed using conventional surface and upper air observation data from 1st January 2013 to 31th December 2015 of Guiyang. Based on different seasons, the number of hidden layers of BP neural network and the node number of hidden layers have been adjusted. By testing the parameters, the model which has the best prediction effect have been selected to carry out in-situ test. The results showed that: the prediction effect in summer was the best, the TS(test score) was 81.6%, the average absolute error was 9.1, the accuracy was 97.4%. Using the current model to validate the prediction effect in summer, it showed that the correlation coefficient of the forecasting value and observed value was 0.71,the average error was 9. The Winter forecast effect was significantly lower than that in other seasons, the TS(test score) was 65.7%, the average absolute error was 19.5, the accuracy was 72.9%. Using the current model to validate the prediction effect in summer, it showed that the correlation coefficient of the forecasting value and observed value was 0.79, the average error was 19. The prediction results of BP neural network model showed no significant relationship with hidden layers and hidden layer nodes.