摘要
基于"分解-集成"的建模思路,构建了变分模态分解(VMD)、弹网惩罚长短期记忆网络(ELSTM)和网格搜索算法(GS)相结合的多模态集成预测模型(VMD-ELSTM-GS),并利用兰州和南京两个城市的PM_(2.5)浓度数据进行了实证预测。研究结果表明:弹网正则化的深度学习方法ELSTM能够提高预测精度,而基于VMD的"分解-集成"范式能够有效降低PM_(2.5)浓度数据的非平稳、高波动程度。
Under the idea of"decomposition-integration",an ensemble prediction model named VMD-ELSTM-GS is proposed,which based on the combination of variational mode decomposition(VMD),long short-term memory network penalized by elastic net(ELSTM)and grid search algorithm(GS).In order to verify the validity of the model,the data of PM2-5 concentrations in Lanzhou and Nanjing city are used to empirically predict-The results show thatH elastic net regularized deep learning method ELSTM can improve the prediction accuracy,while"decomposition-integration"form based on VMD can effectively reduce the degree of non-stationary and high fluctuation of PM2.5 concentration data.
作者
黄恒君
王伟科
HUANG Heng-jun;WANG Wei-ke(School of Statistics,Lanzhou University of Finance and Economics#Lanzhou 730020,China)
出处
《统计学报》
2020年第2期39-47,共9页
Journal of Statistics
基金
国家社会科学基金项目(18BTJ001)
