Research On Air Quality Prediction Method Based On Empirical Modal Decomposition And Area Graph Neural Network

The main pollutant in air quality is PM2.5,and the trend of PM2.5 has complex and nonlinear characteristics.The concentration of PM2.5 is influenced by both temporal and spatial characteristics.Most prediction methods only consider a certain factor that affects PM2.5 unilaterally,while also ignoring the stability characteristics of the sequence.Therefore,this article adopts empirical mode decomposition to transform non-stationary sequences into stationary sequences,uses EMD-XGBoost to analyze temporal dependencies,and uses graph neural network algorithms to learn the spatiotemporal relationships between monitoring stations,achieving accurate prediction of air quality.The main research contents of this article are as follows:Firstly,air quality data set is preprocessed based on feature engineering theory.Correlation analysis was conducted on meteorological features that affect air quality,and corresponding feature selection was achieved.On this basis,the impact of time factors on PM2.5 and the diffusion patterns and interaction principles of PM2.5 between monitoring stations are analyzed,providing a basis for subsequent prediction models.Then,for the air quality data series with time characteristics,a time series analysis method based on EMD-XGBoost is proposed.Input the concentration sequence and meteorological characteristics of the decomposed pollutants into the XGBoost machine learning framework for training and prediction,and explore the evolution law of PM2.5under the influence of complex factors.At the same time,Bayesian optimization is introduced in the process of parameter adjustment to mine the optimal hyperparameter combination,which further improves the analysis effect.On this basis,the advantages and disadvantages of relevant time series prediction methods were compared and analyzed,and comparative experiments were conducted to verify them.Finally,for the air quality data series with temporal and spatial characteristics,a regional map neural network air quality prediction model is proposed.Based on the EMD-XGBoost time series analysis method,temporal correlation features were extracted,and then attention mechanism was used to adaptively adjust the influence weights between monitoring stations to obtain spatial correlation features.At the same time,the influence coefficient adjacency matrix between regions is obtained by fusing the time and space characteristics,which is used as the input of the neural network prediction model of the region map.On this basis,compare the advantages and disadvantages of relevant regional air quality prediction models and conduct experimental verification.Conclusion: The proposed prediction model establishes a regional graph neural network prediction model based on the time series prediction analysis method.This model considers both temporal and spatial correlations,greatly improving the prediction accuracy compared to previous prediction models.