基于Transformer的PM2.5浓度预测方法
A Transformer-based model for PM2.5 concentration prediction
叶耀 1严华1
作者信息
- 1. 四川大学电子信息学院,成都 610065
- 折叠
摘要
在深度学习领域中,通常采用循环神经网络(RNN)等方法对PM2.5的浓度进行预测研究,但传统方法在捕捉多站点数据之间的时空相关性方面存在一定困难.为了解决这一问题,基于Transformer网络模型对PM2.5浓度数据预测进行研究.Transformer采用多头自注意力机制,能够更好地捕捉PM2.5浓度的时空依赖性.其通过模型中编码器提取特征信息,通过模型中解码器处理特征中的依赖关系,输出未来时刻的PM2.5浓度,在真实数据集上的实验表明,Trans-former网络模型具备更好的预测能力.
Abstract
In the field of deep learning,the recurrent neural networks(RNNs)are often used to predict the concentration of PM2.5.However,traditional methods encounter challenges in capturing the spatiotemporal correlations among multi-site data.To ad-dress this issue,research is conducted on predicting PM2.5 concentration using a Transformer-based network model.The Trans-former employs a multi-head self-attention mechanism that better captures the spatiotemporal dependencies of PM2.5 concentration indices across various locations.The model's encoder extracts feature information,while the decoder handles dependencies in the input features to output future PM2.5 concentrations.Experimental results on real datasets demonstrate that the Transformer network model possesses enhanced predictive capabilities.
关键词
PM2.5/Transformer/时空相关性/深度学习Key words
PM2.5/Transformer/spatiotemporal correlation/deep learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据