Jump to Main Content
Estimating contaminant source in chemical industry park using UAV-based monitoring platform, artificial neural network and atmospheric dispersion simulation
- Qiu, Sihang, Chen, Bin, Wang, Rongxiao, Zhu, Zhengqiu, Wang, Yuan, Qiu, Xiaogang
- RSC advances 2017 v.7 no.63 pp. 39726-39738
- Bayesian theory, chemical industry, cloud cover, meteorological data, monitoring, neural networks, unmanned aerial vehicles, China
- Airborne contaminants emitted from chemical industry parks can pose a potential threat to the environment. Therefore, using the data obtained from concentration-monitoring of the contaminant to find the source is of high importance. Most previous source estimation methods collect meteorological parameters and concentration measurements from static sensors. However, some meteorological parameters such as atmospheric stability and cloud cover are difficult to measure precisely. Furthermore, installing only several static sensors does not provide enough sampling data. In this paper, a novel approach is proposed to find the location of an emission source as well as its release rate in a chemical industry park. An unmanned aerial vehicle (UAV) monitoring platform is applied to sample sufficient and high-quality concentration data. Afterwards, an artificial neural network (ANN) trained by an atmospheric dispersion simulation tool is used to locate and quantify the emission source from candidate solutions, bypassing data on the atmospheric stability and other hard-to-obtain meteorological parameters. A numerical simulation with different conditions is implemented to test the accuracy and stability of the proposed approach. A real experiment is conducted in Shanghai to test the performance and sensitivity of this approach as well as the robustness of the monitoring platform. The results show that the approach proposed in this paper can effectively estimate the contaminant source in chemical industry parks. Both the numerical and real experiments prove that the proposed method is less sensitive to errors in meteorological data and concentration measurements than traditional source estimation methods including Bayesian inference and optimization.