Scholarlink

<p>Changes in air quality are closely intertwined with daily life and industrial activities. The research focuses on creating a combined model to forecast the Air Quality Index (AQI) across multiple sites, employing the Maximum Information Coefficient (MIC), Bayesian Optimization Algorithm (BOA), Gaussian Data Augmentation (GDA), and Intensified Spatiotemporal Graph Convolutional Network (ISTGCN). Firstly, the data intensified module was fused on the basis of Temporal Graph Convolutional Network (TGCN) to construct the ISTGCN, and PM2.5 was used as the input of the data intensified module. The PM2.5 data of each station was fed into the model to assist in the prediction of AQI, and the ISTGCN model was constructed. Secondly, the MIC was used to extract the spatial relationship in the multi-site AQI data, and the GDA was used to enhance the original AQI data. GDA increases the total number of samples by adding noise that conforms to the gaussian distribution to the raw data, improving the stability of the model during training. Finally, BOA was employed to optimize the key parameters of ISTGCN to elevate the operation efficiency of the model. In order to evaluate the efficiency of the proposed model, the AQI data of 14 stations in Nanjing were predicted, and the outcomes were benchmarked against eight reference models. The prediction effect of each site is increased by 40% ~ 60%, which realizes the stable prediction of multiple sites.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Simultaneous prediction of air quality for multi-sites using intensified spatiotemporal graph convolutional network based on Gaussian data augmentation

  • 期刊论文

  • 发表日期：

    2026年1月17日

查看全文
  • Xinyu Zhang,
  • Leiming Suo,
  • Changwen Ma,
  • Chu Zhang,
  • Rui He,
  • Muhammad Shahzad Nazir,
  • Tian Peng

摘要

Changes in air quality are closely intertwined with daily life and industrial activities. The research focuses on creating a combined model to forecast the Air Quality Index (AQI) across multiple sites, employing the Maximum Information Coefficient (MIC), Bayesian Optimization Algorithm (BOA), Gaussian Data Augmentation (GDA), and Intensified Spatiotemporal Graph Convolutional Network (ISTGCN). Firstly, the data intensified module was fused on the basis of Temporal Graph Convolutional Network (TGCN) to construct the ISTGCN, and PM2.5 was used as the input of the data intensified module. The PM2.5 data of each station was fed into the model to assist in the prediction of AQI, and the ISTGCN model was constructed. Secondly, the MIC was used to extract the spatial relationship in the multi-site AQI data, and the GDA was used to enhance the original AQI data. GDA increases the total number of samples by adding noise that conforms to the gaussian distribution to the raw data, improving the stability of the model during training. Finally, BOA was employed to optimize the key parameters of ISTGCN to elevate the operation efficiency of the model. In order to evaluate the efficiency of the proposed model, the AQI data of 14 stations in Nanjing were predicted, and the outcomes were benchmarked against eight reference models. The prediction effect of each site is increased by 40% ~ 60%, which realizes the stable prediction of multiple sites.