基于深度信念网络的PM_(2.5)预测  

A prediction method of atmospheric PM_(2.5) based on DBNs

作　　者：郑毅[1] 朱成璋[1] 

机构地区：[1]国防科学技术大学计算机学院,湖南长沙410073

出　　处：《山东大学学报（工学版）》2014年第6期19-25,共7页Journal of Shandong University（Engineering Science）

基　　金：国家高技术研究发展计划(863计划)资助项目(2011AA010702);湖南省科技厅计划资助项目(2012FJ4269)

摘　　要：提出一种基于深度信念网络(deep belief networks,DBNs)的区域PM2.5日均值预测方法,讨论了训练数据选择方式,并优化了DBNs参数设置。通过相关实验并与基于径向基神经网络(radial basis function,RBF)和反向传播神经网络(back propagation,BP)方法比较,验证了基于DBNs方法的可行性和预测精度。实验结果表明:基于DBNs的方法,区域(西安市)预测PM2.5日均值与观测日均值之间均方差(mean square error,MSE)为8.47×10-4mg2/m6;而采用相同数据集,基于RBF和BP的方法均方差为1.30×10-3mg2/m6和1.96×10-3mg2/m6。比较分析表明:基于DBNs的方法能较好预测区域整体PM2.5的日均值变化趋势,显著优于基于神经网络和径向基网络方法的预测结果。A DBNs-based （deep belief networks）method for forecasting the daily average concentrations of PM2. 5 in Xian was proposed.Besides,the way to select training data set as well as the DBNs parameters was optimized.Then relative experiments and comparison with methods based on BP （back propagation）and RBF （radial basis function）ar-tificial neural network confirmed the feasibility and precision of DBNs.The results showed that the MSE （mean square error）between DBNs simulated PM2. 5 daily average concentrations and observed ones was 8. 47 ×10 -4 mg^2/m^6 ,while the MSE of RBF and BP was 1. 30 ×10 ^-3 mg^2/m^6 and 1. 96 ×10 ^-3 mg^2/m^6 respectively.Therefore the DBNs-based method was fit for prediction of PM2. 5 concentrations and it predicted more accurately than those methods based on RBF and BP artificial neural network.

关 键 词：PM2. 5预测 深度信念网络 深度学习 机器学习 限制玻尔兹曼机 

分 类 号：TU457[建筑科学—岩土工程]