An Improved Attention-based BiLSTM for PM2.5 Prediction in Air Environment

The accurate prediction of PM2.5 concentrations in the air is crucial for air quality control and pollution prevention. However, the complexity of air environment data and its time-series characteristics pose significant challenges to this prediction task. This study first employs a random forest algorithm for feature selection to identify key features as input variables. Subsequently, a method combining a Butterworth filter with an attention-based bidirectional long short-term memory (BiLSTM) network is proposed for predicting PM2.5 concentrations. The model smooths the input sequence using a Butterworth filter to remove outliers, and extracts useful information from complex long-term dependencies through a bidirectional long short-term memory (BiLSTM) network based on the attention mechanism. Experimental results demonstrate that this method outperforms other similar approaches.