Research On The Forecast Model Of Graph Convolutional Neural Network For Air Quality

In recent years,the problem of air pollution has become more and more serious.Accurate air quality prediction is of great significance for air pollution control.Air pollution prediction is mainly divided into mechanism model based on atmospheric chemistry model and statistical model based on machine learning.The related technologies based on machine learning have achieved good results in air quality prediction,but there are still several shortcomings as follows :(1)the prediction effect deviation caused by insufficient consideration of spatial correlation.As the pollutants in the air are diffused,the data of the predicted location is affected by the nearby area,which leads to the deviation of the predicted result.(2)The problem of not considering the influence of periodicity of time on the predicted result.Studies have found that air quality has high similarity and correlation at the close time of different periods.If the time period factors between monitoring stations are ignored,the accuracy will be reduced.(3)The linear discrete processing mechanism of time and space discretization is somewhat different from the real situation in the physical world,which leads to certain deviation of the predicted results.In view of the above deficiencies,this paper carries out the corresponding research,the specific content is as follows:First air quality prediction was proposed and the figure of convolution neural network model,the space of the air quality monitoring station information into a graph(graph),the fusion of multiple spatial node information(such as network,in the cities point of interest(poi),air quality monitoring station geographic coordinates and the distance,the forecast data,etc.),and extract the feature space.When the air quality monitoring station information is transformed into graph structure,multiple groups of experiments are used to find the optimal method which is applicable to most data,and the graph structure is transformed into the adjacency matrix of graph convolutional neural network,which solves the problem that the existing air quality prediction model does not consider the spatial correlation sufficiently.Secondly,a spatial-temporal co-occurrence air quality model is proposed,that is,the Attention based on Graph Convolutional Network and Recurrent Unit Neural Network(AGC-GRU).On the basis of spatial relations,integrated time of periodic circulation in traditional neural network to tackle the problem of long distance depend on the gradient to disappear figure convolution model combined with neural network(GCN)and the door control cycle unit(GRU helped),GRU helped operator replacement for convolution operator,the figure characteristics of the processing characteristics of convolution neural network with GRU helped,implements the co-occurrence of time and space,to solve the existing model space is used for discrete linear processing mechanism and traditional circulation neural network model can’t solve the problem of the distance dependence and gradient disappeared.Finally,the PM2.5 and SO2 concentrations of 36 air quality monitoring stations in Beijing were predicted by standard data sets,and the prediction performance of the AGC-GRU model was compared with other existing comparative models.Through the analysis of experimental results,compared with other prediction models,the AGC-GRU model proposed in this paper has the smallest prediction error,higher prediction accuracy and higher reliability.