Research On PM2.5 Concentration Prediction And Space-time Distribution Based On Deep Learning And Satellite Data

In recent years,my country’s economic development has embarked on the fast lane,but the contradiction between environment and development has become more and more prominent.For example,air pollution has been searched for many times.Air pollution such as haze not only has a negative impact on human normal production and life,but also harms people’s body and mind.PM2.5 is the main component of smog,and it is very necessary to predict it scientifically and effectively.When the existing prediction models based on machine learning predict the PM2.5 concentration of a single site,they can often only use the historical data or meteorological data of the target prediction site as the influencing factor for predicting the PM2.5 concentration,and cannot fully consider the target prediction site and its The relationship between the types of surrounding features.In addition,the relationship between the target prediction site and the spatial location of its surrounding monitoring sites is also an important factor.With the flow of air,pollutants in the surrounding area will drift to the area to be measured and be absorbed by vegetation.As an industrialized city in northern my country,the haze problem has been going on for a long time in Harbin.Therefore,a more comprehensive and accurate forecast of Harbin’s PM2.5 concentration and analysis of Harbin’s PM2.5 distribution have become of practical significance subject.This article mainly studies the PM2.5 concentration forecast and the temporal and spatial distribution of PM2.5 concentration in Harbin from the following aspects:(1)Aiming at the problem that traditional methods are difficult to segment remotely sensed images effectively at the same time,and the existing ground object classification methods based on full convolutional neural networks have low classification accuracy in complex scenes,this paper proposes a method based on U-net The improved fully convolutional neural network DL-Unet realizes the effective segmentation of different types of features in remote sensing images.This network improves the traditional convolution method and introduces expanded convolution to increase the receptive field without increasing network parameters.Aiming at the problem of unbalanced feature types in remote sensing images,weighted cross entropy is used as the loss function of the model,which effectively overcomes the selection preference of the model.The adoption of a relatively majority voting strategy for the prediction results further improves the pixel accuracy(PA)of each feature category.Use this model to classify the remote sensing images of the 1km area around the monitoring site,and then analyze the impact of different types of features on the PM2.5 concentration.Quantify the correlation between the four types of ground features and the PM2.5 concentration in preparation for later use as an influencing factor to predict the PM2.5 concentration at the target site.(2)Based on the MODIS data,use the dark pixel method to invert the AOD in the study area,fit the optimal relationship model between AOD and PM2.5 concentration in the four quarters,and use the ArcGis software to map the Harbin area based on the Kriging interpolation method The spatial distribution map of PM2.5 concentration in the four seasons and analysis.(3)Based on the air quality monitoring data and ground weather monitoring data of Harbin City from January 1,2016 to December 31,2017,the changes in PM2.5 concentration from weekly,daily and other time scales were carried out.Analysis;analyze the correlation between PM2.5 concentration and other air pollutants(such as PM 10,NO2,O3,SO2,CO);analyze the impact of atmospheric weather conditions(such as temperature,humidity,wind direction,wind speed,etc.)The influence of PM2.5 concentration;in addition,the temporal and spatial correlation of PM2.5 concentration between the predicted site and the surrounding sites is analyzed.(4)Based on the characteristics of other air pollutants,meteorological factors,the correlation between the stations,and the distribution of ground features around the stations,the prediction model of different PM2.5 concentrations is constructed based on the long and short-term memory neural network(LSTM),and the established The prediction performance of the model is evaluated and compared with other commonly used PM2.5 prediction models.The PM2.5 concentration spatiotemporal prediction model based on pearson correlation analysis proposed in this paper has a higher prediction accuracy for the PM2.5 concentration in Harbin.