An Unsupervised PM2.5 Estimation Method With Different Spatio-Temporal Resolutions Based on KIDW-TCGRU

Abstract

The interpolation of fine-grained air quality has significant prospects in the area of air quality monitoring. The solution to this problem can effectively monitor the air quality of the areas by sparse air quality monitoring stations, so as to reduce the monitoring cost. Most of the existing researches are to solve the problem of air quality monitoring in the areas without stations by different interpolation methods. However, most of them are unable to verify the reliability of the proposed interpolation methods and can't provide a feasible range of interpolation methods at spatial resolutions. Therefore, this paper proposes an effectively unsupervised PM2.5 estimation method based on time distributed convolutional gated recurrent unit (TCGRU) and an interpolation method based on k-nearest neighbor inverse distance weighted (KIDW) to solve these problems. The model is trained by the data obtained by the interpolation method with missing one station in turn to get the monitoring capability of the areas without monitoring stations. Inverse distance weighted is combined with k-nearest neighbor to improve the performance of interpolation. In addition, time distributed convolutional neural network extracts the spatial features of air quality and storage time information for extracting temporal features by the gated recurrent units. A large number of experiments are carried out to evaluate the performance of the method by using the air quality dataset of Hubei Province, China. The experimental results show that the proposed model is effective for the monitoring of PM2.5 in the unsupervised areas and the performance (i.e. MAE=8.23, RMSE=11.27 and MAPE=19.14%) of estimating PM2.5 concentration is better than the comparative methods.