Research on air pollution prediction method based on FE-LSTM

In order to achieve high-precision pollution concentration prediction, this paper proposes a prediction method based on feature-enhanced long short-term memory network (FELSTM). This method introduces a feature enhancement module before the LSTM structure, preprocesses and extracts features of multi-source pollution and meteorological data through onedimensional convolution and attention mechanism, and effectively improves the model’s ability to express nonlinear change trends. The paper first systematically explains the basic principles and feature enhancement mechanism of LSTM, then constructs a FELSTM prediction model, and trains and verifies it in combination with Beijing’s air quality data in 2023. Experimental results show that the model is superior to traditional models and standard LSTM structures in terms of prediction accuracy, stability and regional adaptability, and shows good generalization ability in multi-pollutant modeling and multi-step prediction tasks. The ablation experiment further verifies the key role of the feature enhancement module in improving model performance. This study can provide technical support for the intelligent prediction system of urban air pollution, and has important application value for environmental supervision and public health protection.