基于混合深度神经网络的大气污染预测  

Prediction of Air Pollution Based on Mixed Deep Neural Networks

作　　者：宋耀宇 Song Yaoyu(North China University of Water Resources and Electric Power,Zhengzhou Henan 450046,China)

机构地区：[1]华北水利水电大学

出　　处：《信息与电脑》2019年第24期99-101,共3页Information & Computer

摘　　要：城市空气污染物分布受各种因素的影响,准确预测空气污染物分布情况可以提高环境管理的决策能力,防止严重空气污染事件的发生。基于大量的气象数据和深度学习技术,笔者提出了一种新的大气污染物浓度预测方法。该方法使用卷积神经网络作为基础层的设计,可自动提取输入数据的特征。输出层运用一个长短时记忆网络,以保证时间依赖性。通过性能优化,该模型可以按时间序列预测未来的细颗粒物(PM2.5)质量浓度。使用郑州市的历史气象数据,将模型预测结果与实际结果进行比较;实验结果表明,与经典深度学习模型相比,该算法提高了预测性能。The distribution of urban air pollutants is constantly changing under the influence of various factors.Accurate prediction of the distribution of air pollutants can improve the decision-making ability of environmental management and prevent the occurrence of serious air pollution events.Based on a large number of meteorological data and deep learning technology,the author proposes a new prediction method for atmospheric pollutant concentration.The method use convolutional neural network as the basic layer design and can automatically extract the characteristics of the input data.A long short-term memory network is used for the output layer to ensure time dependence.Through performance optimization,the model can predict the future PM2.5 concentration according to time series.Using the historical meteorological data of Zhengzhou city,the prediction results of the model are compared with the actual results.Experimental results show that compared with the classical deep learning model,this algorithm improves the prediction performance.

关 键 词：混合深度神经网络 大气污染物浓度预测 细颗粒物(PM2.5) 

分 类 号：X51[环境科学与工程—环境工程]