SFDformer: A Frequency-Based Sparse Decomposition Transformer for Air Pollution Time Series Prediction

Zhenkai Qin ^1,2 Baozhong Wei ^2* Caifeng Gao ^2* Xiaolong Chen ^3* Hongfeng Zhang ^3* Cora Un ^3*

• ¹ Southwest Jiaotong University, Chengdu, China
• ² Guangxi Police College, Nanning, Guangxi Zhuang Region, China
• ³ Faculty of Humanities and Social Sciences, Macao Polytechnic University, Macao, Macau Region, China

With the rapid advancement of industrialization and the prevalent occurrence of haze weather, P M 2.5 contamination has emerged as a significant threat to public health and environmental sustainability. The concentration of P M 2.5 exhibits intricate dynamic attributes and is profoundly correlated with meteorological conditions as well as the concentrations of other pollutants, thereby substantially augmenting the complexity of predictive endeavors. Consequently, a novel predictive methodology has been developed, integrating time series frequency domain analysis with the decomposition of deep learning models. This approach facilitates the capture of interdependencies among high-dimensional features through time series decomposition, employs Fourier Transform to mitigate noise interference, and incorporates sparse attention mechanisms to selectively filter critical frequency components, thereby enhancing time-dependent modeling.Importantly, this technique effectively reduces computational complexity from O(L 2 ) to O(L log L).Empirical findings substantiate that this methodology yields notably superior predictive accuracy relative to conventional models across a diverse array of real-world datasets. This advancement not only offers an efficacious resolution for P M 2.5 prediction tasks but also paves the way for innovative research and application prospects in the realm of complex time series modeling.