摘要
为提高城市空气质量预报准确率,文章在传统BP神经网络的基础上提出了基于气象相似准则的样本优化方法,建立了三层样本筛选优化机制,确定了阀值及权重矩阵,从而建立了城市空气质量动态预报模型。将模型应用在广州8个空气质量监测站点的预报上,并与传统的BP神经网络空气质量预报模型进行了对比分析,效果良好。分析结果表明,广州8个空气质量监测站点的SO2、NO2、PM10/2.5的实测值与预报值的平均绝对误差分别为0.016 mg/m3、0.014 mg/m3、0.020 mg/m3,级别预报准确性评分分别为89.6、92.6和84.6,预报准确度综合评分达81.6,并且比传统神经网络模型具有更高的预报精度。
A simple optimized method based on the meteorological similarity criteria and the traditional BP neural network was proposed in this paper.Through setting up an optimization mechanism of three-tiered sample screening as well as the threshold and weighing matrix,a dynamic forecast model for urban air quality was established.Applications of this model to forecasting air quality with respect to SO2,NO2 and PM10/2.5 in eight monitoring stations of Guangzhou City showed the better results in terms of prediction accuracy than those using traditional BP neural network model,with a general mark of 81.6 for prediction accuracy.
出处
《环境科学与技术》
CAS
CSCD
北大核心
2013年第5期156-161,共6页
Environmental Science & Technology
基金
国家科技支撑计划项目(2011BAG07B00)
国家自然基金项目(51108471)
广东省自然科学基金(S2011040002839)
关键词
空气质量动态预报
气象相似准则
样本优化
BP神经网络
air pollution dynamic forecast
meteorological similarity criteria
sample optimization
BP neural network
