Research On AQI Evolution Prediction Model By Fusing GCN And LSTM

With the rapid industrialization of China and the gradual improvement of the quality of life of the nation’s residents,people’s living environment requirements are getting higher and higher,and they are more and more concerned about air quality issues.The current research on air quality focuses on two aspects:on the one hand,the analysis of spatial and temporal heterogeneity of air quality,and on the other hand,the construction of air pollution prediction models.Most of the existing air quality spatial and temporal heterogeneity analyses are based on the inverse data from remote sensing images as the data source to explore the spatial distribution and influencing factors of a specific pollutant,but the determination of the primary pollutant in such studies is based on global analysis without considering spatial differences.Although remote sensing data can achieve full regional coverage,cloud obscuration sometimes occurs and excessive noise,which makes it difficult to realize the study of heterogeneity in different time dimensions.Heterogeneity study.At the same time,the accuracy of the inverse performance of this method is difficult to guarantee,and ground station data are needed to supplement the correction.Most of the existing air quality predictions are made for time-series data without considering the spatial relationship between multiple stations,resulting in data features that do not contain spatial relationships,and the final trained model does not predict the corresponding target values more accurately.To address the above issues,the thesis focuses on the Yangtze River Delta as the study area and conducts the following research.1)Spatial and temporal heterogeneity of air quality in the Yangtze River Delta.By pre-processing the collected air quality data,meteorological data and vegetation cover data from 2016 to 2018 in the Yangtze River Delta region with missing values and singular values;using spatial correlation analysis to analyze the spatial and temporal characteristics of AQI and its aggregation in the study area;using gray correlation analysis to calculate the relationship between six basic pollutants(PM2.5,PM1O,SO2,NO2,O3,CO)and AQI,and analyzed the differences of the top pollutants in different regions of the study area;the spatial differences of various influencing factors behind AQI were analyzed by using geographically weighted regression to provide corresponding strategies and preventive decision-making suggestions for regional differentiated air quality improvement.2)Modeling air quality index prediction model based on deep learning method.The six basic pollutants of air quality,meteorological data(temperature,wind speed,precipitation,relative humidity,and barometric pressure)and vegetation cover index of the Yangtze River Delta region are used as inputs,and AQI is used as output.Taking Nanjing city as an example,the prediction models based on LSTM and GCN-LSTM for AQI values were constructed and the accuracy was evaluated.To prove the generalizability of the model,the predictions were carried out in Hangzhou,Hefei and Shanghai,and the results showed that the GCN-LSTM model still outperformed the LSTM model in terms of prediction accuracy in these three cities,and the model has strong generalizability.In the study of spatial and temporal heterogeneity of air quality,the AQI showed an obvious U-shaped feature in the temporal dimension with higher values in summer and lower values in winter,and in the spatial dimension,the AQI index was higher in the northwestern part of the study area and lower in the southeastern part.The gray correlation analysis shows that PM2.5 and PM10 are basically the two highest correlations among the 41 cities in the study area and belong to the top pollutants in the region.The results of the geographically weighted regression revealed that the influence of relative humidity on AQI was the greatest and the vegetation cover index was the least influential.In the air quality model prediction,using Nanjing as the experimental site,the results show that the RMSE of the LSTM model is 18.56,MAE is 15.08,MAPE is 28.89%,and R2 is 71.19%,while the GCN-LSTM model is 14.80,8.78,15.81%,and 88.33%,respectively,and the results prove that the GCN-LSTM network is more accurate than the LSTM network can make air quality prediction more accurately.The experimental results of the model validation in Shanghai,Hangzhou and Hefei demonstrate that the GCN-LSTM model outperforms the LSTM in prediction.