APP下载

基于交互差分时空LSTM的网格化臭氧浓度预测

2023-07-13刘恩海任晓康张智李妍赵娜张军

河北工业大学学报 2023年3期
关键词:信息模型

刘恩海 任晓康 张智 李妍 赵娜 张军

摘要 臭氧濃度的预测对于大气环境治理、空气质量改善等起到了重要的作用。本文提出了一种交互差分时空LSTM网络预测模型(ST-IDN)来挖掘臭氧浓度历史数据的时间相关性和空间相关性,并成功将其应用到网格化臭氧浓度数据预测上。在该模型中,首先交互模块(IC)可以通过一系列的卷积操作来捕捉短期上下文信息,其次层融合模块(LF)可以融合不同层的空间信息来获得上一时刻丰富的空间信息,最后差分时空LSTM模块(DSTM)将捕捉到的时间信息和空间信息进行统一建模实现臭氧浓度预测。所构建模型分别与卷积LSTM网络(ConvLSTM)、预测循环神经网络(PredRNN)以及Memory in Memory网络(MIM)模型在河北省气象局提供的臭氧浓度数据上进行了对比分析, ST-IDN模型的平均绝对误差分别降低了19.836%、12.924%、7.506%。实验结果表明,所提出的模型能够提高臭氧浓度的预测精度。

关 键 词 交互差分时空LSTM预测网络;网格化臭氧浓度数据;臭氧浓度预测;时间信息;空间信息

中图分类号 TP391     文献标志码 A

文章编号:1007-2373(2023)03-0036-08

DOI:10.14081/j.cnki.hgdxb.2023.03.004

Prediction of gridded ozone concentration based on interaction differential spatiotemporal LSTM

LIU Enhai1, REN Xiaokang1, ZHANG Zhi2, LI Yan1, ZHAO Na2, ZHANG Jun1

(1. School of Artificial Intelligence, Hebei University of Technology, Tianjin 300401, China; 2. Hebei Meteorological Disaster Prevention and Environmental Meteorology Center, Shijiazhuang Hebei 050000, China)

Abstract The prediction of ozone concentration plays an important role in controlling atmospheric environment and improving air quality. In this paper, an interactive difference spatiotemporal LSTM network prediction model (ST-IDN) is proposed to mine the temporal and spatial correlation of the historical data of ozone concentration, and is successfully applied to the prediction of gridded ozone concentration data. In this model, firstly, the interactive module (IC) can capture short-term context information through a series of convolution operations, and secondly, the layer fusion module (LF) can fuse spatial information of different layers to obtain abundant spatial information at the last moment. Finally, the differential space-time LSTM module (DSTM) will use the captured time and space information for unified modeling to predict ozone concentration. The constructed models were compared with convolution LSTM network (ConvLSTM), predictive cyclic neural network (PredRNN) and Memory in Memory network (MIM) models respectively on the ozone concentration data provided by Hebei Meteorological Bureau. The average absolute error of ST-IDN model is reduced by 19.836%, 12.924% and 7.506%, respectively. The experimental results show that the proposed model can improve the prediction accuracy of ozone concentration.

Key words interaction differential spatiotemporal LSTM prediction network; gridded ozone concentration data; prediction of ozone concentration; time information; spatial information

0 引言

随着空气城市化进程的不断加快以及工业的快速发展,臭氧污染越来越严重[1-2],臭氧污染不仅会对人类的身体健康构成严重威胁,还会对工农业生产造成一定的影响[3-4]。通过对臭氧浓度的预测,可以帮助人们知晓未来臭氧浓度变化情况,提前做好防护。因此,准确地预报臭氧浓度变化是十分必要的。

当前臭氧预测的方法大致可以分为3类:数值模型预测算法、统计模型预测算法和深度学习预测算法。……

登录APP查看全文

猜你喜欢

信息模型
一半模型
重尾非线性自回归模型自加权M-估计的渐近分布
订阅信息
3D打印中的模型分割与打包
FLUKA几何模型到CAD几何模型转换方法初步研究
展会信息
一个相似模型的应用
信息
健康信息
健康信息(九则)