Implementation of air pollution traceability method based on IF-GNN-FC model with multiple-source data

The most difficult problem in air pollution control is the accurate traceability of pollutants because atmospheric motion is affected by many factors. In this paper, based on the analysis of data, we propose an isolated forest-graph neural network-full connectivity algorithm, i.e., the IGF model to avoid the problem of capturing spatial information between data points due to the presence of unfixed relationships between data points in multi-source air pollution data, which cannot be represented through regular tensors or matrices. We utilize the IGF model to analyze the distribution characteristics of hourly air quality data and meteorological data in eight regions of Beijing between January 1, 2022, and May 31, 2022. First, the feature importance of each region is predicted, which is used to calculate the contribution of each region, and then, the pollution sources are divided into regional classes based on the regional contribution, and the spatial distributions of stations of different classes are combined to determine the final traceability results. The final traceability results are determined by combining the spatial distribution of sites of different levels. The experimental results show that the IGF model has higher prediction accuracy than the mainstream models to ensure the accuracy of traceability. Combining regional traffic flow, energy consumption, and geographical factors, we acquire the result that Chaoyang District has the highest level of pollution, with O 3 as the main pollutant, and Changping and Huairou Districts have the lowest levels of pollution. This result can provide a scientific basis and technical support for the government to control air pollution in Beijing.