Scholarlink

<p>Locating atmospheric pollution sources is essential for protecting the environment and public health. It enables identification and control of pollution sources, reduces the impact of pollutant emissions on the environment, and holds significant practical importance. This study proposes a method that integrates sensor networks with machine learning to identify pollution sources. By monitoring sensor data, we track pollutant dispersion and apply models including support vector machines (SVM), deep neural networks (DNN), and hierarchical DNNs (HI-DNN) for source classification. To overcome the limitations of numerical simulations, we collected real-world diffusion data through field measurements. We partitioned the monitoring area into grids and transformed the localization task into a multi-class classification problem. We trained and evaluated models based on DNN, SVM, and HI-DNN architectures. The SVM model with an RBF kernel achieved an accuracy of 89.5%, the basic DNN model reached 95.2%, and the HI-DNN model achieved 96.7%, successfully identifying the most likely source among 17 candidates. Most misclassified predictions were near the actual source, providing useful reference for practical localization.</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

A Method for Locating Air Pollution Sources Combining Sensor Network and Machine Learning

  • 期刊论文

  • 发表日期：

    2026年1月20日

查看全文
  • Juntao Hu,
  • Can Cui,
  • Jingjing Gao,
  • Ao Zhang,
  • Xinghua Liu,
  • Shicheng Zhang,
  • Yong Fang

摘要

Locating atmospheric pollution sources is essential for protecting the environment and public health. It enables identification and control of pollution sources, reduces the impact of pollutant emissions on the environment, and holds significant practical importance. This study proposes a method that integrates sensor networks with machine learning to identify pollution sources. By monitoring sensor data, we track pollutant dispersion and apply models including support vector machines (SVM), deep neural networks (DNN), and hierarchical DNNs (HI-DNN) for source classification. To overcome the limitations of numerical simulations, we collected real-world diffusion data through field measurements. We partitioned the monitoring area into grids and transformed the localization task into a multi-class classification problem. We trained and evaluated models based on DNN, SVM, and HI-DNN architectures. The SVM model with an RBF kernel achieved an accuracy of 89.5%, the basic DNN model reached 95.2%, and the HI-DNN model achieved 96.7%, successfully identifying the most likely source among 17 candidates. Most misclassified predictions were near the actual source, providing useful reference for practical localization.