摘要
本文提出一种融合时空特征的城市多站点PM2.5预测方法,该方法可以捕捉PM2.5在时间和空间上的相关性,通过将区域多个站点的PM2.5数据转换为一系列静态图像,将其输入到卷积长短期记忆(ConvLSTM)模型中,采用端对端的方式进行训练,预测城市未来多个站点多个时段的PM2.5浓度。以北京多个站点的PM2.5数据进行实验验证。结果表明:考虑了时空特征的ConvLSTM方法在短期预测方面优于其他4种时序方法,该方法可为PM2.5预测提供新的思路。
An urban multi-site PM2.5 prediction method that fuses spatial-temporal features is proposed,which can capture the temporal and spatial correlation of PM2.5.By converting regional PM2.5 data of multi-site into a series of static images,and input into the ConvLSTM model.The model is trained in an end-to-end manner to predict the PM2.5 concentrations of multiple stations and periods in the future.The method is verified by the experiment with the PM2.5 data of multi-site in Beijing.The results show that ConvLSTM method considering spatial-temporal features is better than the other 4 timing methods in short-term prediction,which provides a new idea for PM2.5 prediction.
作者
黄琨
吴学群
成飞飞
韩啸
HUANG Kun;WU Xuequn;CHENG Feifei;HAN Xiao(Faculty of Land Resource Engineering,Kunming University of Science and Technology,Kunming 650093,China)
出处
《传感器与微系统》
CSCD
北大核心
2024年第5期149-152,157,共5页
Transducer and Microsystem Technologies
关键词
时空特征
卷积长短期记忆
多站点
PM2.5浓度预测
spatial-temporal features
convolutional long short-term memory(ConvLSTM)
multi-site
PM2.5 concentration prediction