Frequency-Aware Attention-LSTM for PM$_{2.5}$ Time Series Forecasting
Jiahui Lu, Shuang Wu, Zhenkai Qin, Guifang Yang
TL;DR
This work tackles the challenge of forecasting $PM_{2.5}$ concentrations in urban environments by integrating frequency-aware preprocessing with deep temporal modeling. The proposed FALNet architecture applies STL decomposition and FFT-based denoising to separate and clean the signal, then uses a two-layer LSTM to capture temporal dependencies, followed by a multi-head attention mechanism to refine predictions with focus on salient time steps. The approach demonstrates improved predictive accuracy (lower MAE/RMSE and higher $R^2$) and robustness to sharp pollution spikes on real-world urban data, highlighting its potential for real-time air quality monitoring and decision support. The authors also discuss limitations such as occasional overfitting in long low-concentration regimes and suggest future work to incorporate meteorological, spatial, and graph-based information to enhance generalization and cross-regional applicability.
Abstract
To enhance the accuracy and robustness of PM$_{2.5}$ concentration forecasting, this paper introduces FALNet, a Frequency-Aware LSTM Network that integrates frequency-domain decomposition, temporal modeling, and attention-based refinement. The model first applies STL and FFT to extract trend, seasonal, and denoised residual components, effectively filtering out high-frequency noise. The filtered residuals are then fed into a stacked LSTM to capture long-term dependencies, followed by a multi-head attention mechanism that dynamically focuses on key time steps. Experiments conducted on real-world urban air quality datasets demonstrate that FALNet consistently outperforms conventional models across standard metrics such as MAE, RMSE, and $R^2$. The model shows strong adaptability in capturing sharp fluctuations during pollution peaks and non-stationary conditions. These results validate the effectiveness and generalizability of FALNet for real-time air pollution prediction, environmental risk assessment, and decision-making support.