Research On Multi-layer Attentional Neural Network Model For Air Quality Inference

Air pollution can cause serious harm to humans and ecosystems,and spatially accurate air quality information can help people avoid unhealthy outdoor activities.However,due to high construction and maintenance costs,the number of monitoring sensors deployed is limited,and fine-grained air quality inference in space is a difficult task.Most of the existing research only concerns the spatial area where the sensor has been deployed(prediction,real-time display,analysis,etc.).Due to the limitations of actual deployment of sensors,the granularity of the sensor coverage is usually relatively coarse,which has a better effect on nearby areas,but for areas far away from the deployment site,it cannot truly reflect the actual situation.At present,the fine-particle air quality in the inferred space is mainly divided into two categories: physical methods and data-driven methods.The existing methods have some shortcomings:(1)The influence of the neighboring area of the area to be inferred is not considered.(2)It does not take into account that the importance of each monitoring point for air quality inference is different at different times.(3)For time series data,the impact of different historical time periods on air quality has not been considered.In response to the above problems,the research content of this article is as follows:First of all,in view of the problem of data loss in the sensor deployment area,this paper uses the stacking model fusion method to integrate the three strong learners of Random Forest,XGBoost and Light GBM.Finally,the missing values in the data set are filled to improve the performance of the subsequent multi-layer attention air quality inference model.Second,this article divides the city into grid areas of the same size.The air quality of each grid area is affected by the adjacent grid area.In order to infer the accurate air quality in areas where air quality monitoring sensors are not deployed,this paper proposes an air quality inference neural network model based on multi-level attention mechanism spatial particle air quality inference(MAI).Using time series data and nonlinear data of the area to be inferred,the grid area around the area and the deployed sensor area are inferred.Time series data uses a multi-layer attention mechanism to adaptively assign different weights to different locations,grid areas and historical time slice data to improve the accuracy of the air quality inference of the model.Finally,experiments are conducted on the public air quality data set.Compared with many current models with the best performance,the MAI model proposed in this paper has higher accuracy.