基于神经网络的小尺度环境下生物气溶胶溯源方法研究.

岳 超; 杜耀华; 张 宁; 薛 腾

Objective To propose a biological aerosol traceback method based on neural network to accurately obtain sourceinformation in events such as bioterrorist attacks and biological spills in small-scale environments. Methods Firstly, computational fluid dynamics (CFD) was used to simulate the diffusion state of a released biological aerosol ina small-scale environment, and the distribution characteristics of biological particles at different points in the release area were analyzed. Secondly, the release sources and monitoring points were reasonably laid out to simulate a biological pollution event and experimentswere conducted on the release process of bioaerosols to collect the concentration information, location information andmeteorological information at the release points and monitoring points. Finally, traceability models were built using artificialneural networks with different structures of AlexNet, ResNet18 and ResNet34. There were 85% of the simulation datagenerated from CFD simulation were used for model training and 15% for model testing, and the optimal model was selectedand validated by in situ experiments. Results ResNet18 model behaved better than AlexNet and ResNet34 models in easycomputation and low time consumption with guaranteed accuracy. Under natural diffusion conditions ResNet18 model couldeffectively predict the concentration, longitude and latitude of the release source with relative error rates of 3.46%, 4.87% and4.98%, respectively. Conclusion The neural network -based bioaerosol traceability method has the advantages of highaccuracy and low time consumption, which can accurately reflect the release source location and concentration information ofbiological aerosols in small-scale environments.

Chinese Medical Equipment Journal, 2023, Vol 44, Issue 2, p16

1003-8868

Academic Journal

10.19745/j.1003-8868.2023024