Optimizing Real-Time PM2.5 Predictions with Deep Belief Networks: Synergistic Role of Kriging Interpolation

Read full text Summarise with VIRAT
Authors: Jin Xiao, Chao Bian, Guangqiu Huang
Cited by: 0
References: 19

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The aim of this study was to improve the accuracy and stability of fine particulate matter (PM2.5) concentration prediction by integrating a deep belief network (DBN) and the kriging interpolation algorithm. First, we utilized a binary particle swarm optimization algorithm to construct the evaluation model, thereby identifying the main factors that most significantly influence the prediction of the PM2.5 concentration. Second, we thoroughly analyzed the relationship between the real-time PM2.5 concentration in Xi’an City during different months (or days) and various traditional variograms to obtain training samples for DBN. Finally, on the basis of the selected functions and sample data, we obtained corresponding kriging model interpolation results. The experimental results showed that, compared with the prediction errors of the traditional, spatiotemporal, and empirical kriging interpolation methods, the proposed method yielded a prediction error of only 4.677%. This error was calculated using the Mean Absolute Percentage Error (MAPE), which reflects the average percentage deviation between predicted and actual PM2.5 concentrations. MAPE was chosen to facilitate comparison with previous studies in air quality prediction. This method increases not only the accuracy of the interpolation algorithm but also its robustness. The core contribution of this study is the introduction of a DBN to optimize the kriging interpolation algorithm, significantly improving the accuracy and stability of PM2.5 concentration prediction. We combined climate and meteorological data with a deep learning technique to enable the model to adaptively select the most appropriate variogram, thereby more accurately simulating the air quality during different seasons and in various environments. In summary, this study provides a new and effective method for accurate air quality simulation, which has significant implications for advancing meteorological research and improving environmental pollution conditions.

Fields of Study & Subjects

Air Quality Monitoring and Forecasting Air Quality and Health Impacts Traffic Prediction and Management Techniques

Publication details

Journal / source
Polish Journal of Environmental Studies
Volume
35
Issue
5
Pages
6447–6462
Publication date
26 March 2026
Language
EN