A new method for predicting PM2.5 concentrations in subway stations based on a multiscale adaptive noise reduction transformer -BiGRU model and an error correction method

Abstract

PM2.5 is a significant contributor to air pollution, with a notable impact on human health. Subway stations, with their high pedestrian traffic, present a particular challenge in this regard. By monitoring PM2.5 levels, subway managers can take prompt action, such as optimizing the operation of air purification equipment in stations, to enhance air quality within stations and thereby enhance the passenger experience. This paper proposes an enhanced Transformer-BiGRU prediction model, which incorporates a MSHAM(Multiscale Hybrid Attention Mechanism)comprising a multi-scale convolutional attention mechanism and a VMD decomposition self-attention mechanism. Additionally, a ANR(Adaptive Noise Reduction) module has been integrated into the model to facilitate noise reduction. Finally, the prediction is performed by BiGRU. The resulting error sequence is predicted by BiGRU and the predicted sequence is corrected. In this paper, a dataset of pollutants from Seoul subway stations in South Korea is used to compare with the base model. The model presented in this paper achieves the highest accuracy.