FORECASTING PARTICULATE MATTER CONCENTRATIONS WITH DEEP NEURAL NETWORKS USING METEOROLOGICAL DATA

Le résumé fourni par la source

Particulate matter (PM), especially PM10 and PM2.5, poses a significant threat to human health and the environment. Accurate forecasting of particulate matter concentrations is essential for air quality management, public health protection, and the implementation of effective pollution reduction strategies. In this study, we investigate the problem of short-term forecasting of PM10 and PM2.5 concentrations using historical measurements. To address this challenge, we propose and evaluate two independent deep learning approaches: the Kolmogorov–Arnold Network (KAN) and the temporal fusion transformer (TFT). The KAN model is designed to capture complex nonlinear relationships within air quality time series, whereas the TFT architecture utilises attention mechanisms to model temporal dependencies and identify relevant patterns over time. The proposed models are trained and tested on real-world air quality datasets in two ways: using only historical concentrations and using both historical concentrations and meteorological data. The forecasting performance is assessed using standard metrics, including MAE, MSE, and R2 measures. Experimental results confirm the efficacy of the proposed methods, compared with the LSTM-based baseline model.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.