In recent years, the frequency of grey weather in North China has increased, which has seriously affected people's normal life. The increase of the concentration of atmospheric fine particles (PM2.5) is a landmark feature, and the harm of atmospheric PM2.5 pollution to human health has attracted more and more attention. High concentration of PM2.5 will not only lead to the decline of lung function, but also increase the incidence and mortality of respiratory diseases, and even affect people engaged in outdoor physical exercise. Firstly, this paper introduces the pollution status and harm of fine particles, as well as the technology and equipment adopted by the state for its monitoring at this stage, which has accumulated some prior knowledge for the establishment of soft sensing model. This paper systematically introduces the structure and learning process of BP and RBF neural networks, and determines the feasibility of establishing a soft-sensing model based on BP and RBF neural networks to conduct soft-sensing experiments on the concentration of fine particles in the atmosphere. The results show that the chemical component that contributes the most to the mass concentration of PM2.5 is nss-SO42-, accounting for 27.2%, followed by OM and NH4+, while the concentration of NO3- is low, accounting for only 4.4%. Generally speaking, per capita gas pollutant emissions decrease with the increase of per capita income. The estimated value of the proportion of the tertiary industry is sometimes not significant, but as long as it is significant, its symbols are all negative, which is in line with expectations, indicating that developing the tertiary industry is conducive to improving the environmental quality.