I. Introduction
With the continuous development of urbanization and industrialization, the problem of atmospheric pollution has become an important part of the global environmental problem. CO is an important indicator of environmental pollution, which has an important impact on human health and environmental protection. Some scholars at home and abroad use the data of air workstations to predict the concentration of air pollutants. Researchers and experts use different models and methods to predict atmospheric pollution problems, including deep learning methods such as 3D unsteady Lagrange diffusion model [1], Gaussian plume model [2], CNN [4], LSTM [3], and CNN-BILSTM [5]. These models have high accuracy in predicting pollutant concentrations, but deviations occur when pollution and meteorological factors change. In order to solve the problem of limited data samples and time series dependence, researchers have combined different models and methods and used optimization algorithms to select hyperparameters to improve the performance and ability of prediction models. In recent years, swarm intelligence algorithms such as sparrow search optimization algorithm have also been proposed to optimize the hyperparameters of the model to further improve the adaptation performance of the model.