Research On The Evolution Mechanism Of Air Pollution Based On Time Series Network

Environmental air pollution has become a major environmental challenge facing China and the world,which has serious health and socio-economic consequences.Studying the dynamic characteristics of the time series of atmospheric pollutant concentration and exploring the fluctuation trend of atmospheric pollutant concentration plays an important role in revealing the changing law of air pollutants,controlling air pollution and formulating pollution control policies.Based on the theories and methods related to complex networks,this paper studies the dynamic characteristics of atmospheric pollutant concentration fluctuations in Beijing based on the time series of six types of atmospheric pollutant concentrations from January 2015 to December2019.The detailed research contents are as follows.（1）The False Nearest Neighbours method and C-C algorithm were used to estimate the time delay and embedded dimension in the phase space reconstruction process,and embed the atmospheric pollutant concentration time series into the high-dimensional phase space.The phase points in phase space were regarded as nodes of the recurrent network,and the existence of edges was determined by the comparison of the distance between phase points and critical threshold.Through the variation of degree distribution,density and clustering coefficient changes with the critical threshold,the optimal critical threshold of constructing recurrent network was determined.（2）Complex networks were constructed by recurrent plot and visibility graph algorithm.The similarities and differences of various networks were compared and analyzed.The results showed that the dynamic characteristics of the air pollutant concentration time series were reflected by the network structure and the adjacency matrix diagram.The degree distribution of the recurrent network of different air pollutant concentrations obeyed the distribution of Gaussus or power-law,and there were obvious differences between network parameters,reflecting the dynamic characteristics of different pollutant concentrations.The degree distribution of different pollutant concentrations’visibility graph obeyed the approximately same power-law distribution,the networks had typical scale-free characteristics,and the characteristic parameters were at the same level.（3）A transition networks model of multi-atmosphere pollutant concentration fluctuation time series was established,and the co-evolution characteristics of time series were analyzed.The time series of PM_2.5,PM₁₀,NO₂ and O₃ concentrations were coarse-grained and symbolized.The concentration fluctuation symbol combination（fluctuation patterns）was regarded as the node of the transition network,the conversion between nodes represented the edge,and the edge weight was the frequency of the transition between patterns.A weighted and directed conversion network model was constructed.The characteristic parameters of the network represented information such as the dynamic characteristics of the pollutant concentration fluctuation pattern.The degree and intensity distribution of nodes obeyed the power-law distribution,and the betweenness centrality of the nodes was quite different.The pattern that characterized small fluctuations in the concentration of air pollutants was considered to have an important connection effect.The average shortest path,diameter,and close centrality of nodes were used to measure the speed and direction of the transition between fluctuation patterns to a certain extent.According to the number and direction of the edges connecting the nodes,a probability matrix predicting concentration fluctuations was constructed to predict short-term atmospheric pollutant concentration fluctuations.