PM10 Time Series Forecasting Using Variational Mode Decomposition Informer

Abstract Accurate forecasting of particulate matter (PM) is crucial for environmental protection and human health. This study proposes a hybrid prediction model to forecast PM concentrations with an aerodynamic diameter of less than 10 µ m (PM10) at a concrete batching plant (CBP) involved in highway construction. This model combines variational mode decomposition (VMD) and an informer, also referred to as VMD-informer. The VMD is employed to denoise the PM10 data, while additional factors, including temperature, humidity, ambient noise, light intensity, and PM2.5, are integrated to predict PM10 concentrations using the informer method. The effectiveness of the model was confirmed using data from a CBP associated with a national highway. The results showed that the VMD-informer model outperformed the informer-only model, enhancing the accuracy of the time series forecasting (TSF) of PM10 concentrations by 39.9% and 23.7% in the mean-square error (MSE) and mean absolute error (MAE), respectively. Further, the model demonstrated improved efficiency by reducing attention computation and memory consumption. Consequently, the proposed model is suitable for PM10 TSF in CBPs involved in highway construction. This advancement supports the implementation of hygiene protection measures and reduces environmental pollution.