Research On Prediction Method Of PM2.5 Concentration Based On Spatiotemporal Convolutional Network

With the rapid development of urbanization and urban industrialization in China,air pollution has become increasingly serious,which has negative impacts on human health,social environment,and national economic development.PM2.5 is an important indicator for measuring air pollution levels,and monitoring and predicting its concentration can provide effective reference for air quality management and decision-making.However,existing PM2.5concentration prediction models are often complex,expensive,and have poor predictive accuracy,making it difficult to achieve optimal applications in practical environments.Moreover,most current research only uses historical atmospheric pollution and meteorological data of the target station to predict PM2.5 concentration,while ignoring the spatiotemporal correlation between the target station and nearby monitoring stations.Therefore,in order to improve the accuracy and reliability of PM2.5 concentration prediction for urban target stations,this paper proposes a method of utilizing spatiotemporal convolutional neural network algorithm for prediction.Specifically,the main research workflow and conclusions in this paper can be summarized as follows:(1)With the aim of demonstrating the necessity and feasibility of predicting PM2.5concentration,this paper firstly analyzes and discusses the concept,sources,formation process,hazards,and related background knowledge of PM2.5.Secondly,by reviewing the current research status in the field of PM2.5 concentration prediction both domestically and internationally,the shortcomings of existing prediction models are discussed,and the superiority of spatiotemporal convolutional networks in this field is emphasized,laying the foundation for subsequent research and work.Moreover,by introducing the relevant theoretical methods for PM2.5 concentration prediction,this paper provides technical support and theoretical basis for the construction of prediction network models.(2)To obtain the temporal and spatial characteristics of the target station and multiple monitoring stations,this paper conducted temporal correlation analysis and spatial correlation analysis of PM2.5 concentration.Firstly,the collected dataset was cleaned to ensure data integrity,and the temporal trend of PM2.5 concentration at the target station was studied.Secondly,Pearson correlation coefficient was used to analyze the temporal correlation between the PM2.5 concentration of the target station and other influencing factors at the station.In addition,the maximum information coefficient was used to analyze the spatial correlation between the PM2.5 concentration at the target station and other monitoring stations,providing technical support for establishing effective spatiotemporal prediction models.(3)To address the issues of model complexity,high computational cost,and poor prediction accuracy in existing convolutional neural network-based models,this paper proposes the use of a temporal convolutional network to process and predict PM2.5 concentration at the target station.This temporal model has advantages such as a simple architecture,easy interpretability,high accuracy,and low computational cost.It can effectively extract data features through dilated convolutions and capture long-term temporal dependencies of the data,which is an effective method to improve the prediction accuracy of PM2.5 concentration and reduce model training complexity.By using evaluation metrics such as MAE,RMSE,and R~2and comparing with common convolutional network models,the performance of proposed model was tested and validated,demonstrating its effectiveness and superiority.This work lays the foundation for future research on improved PM2.5 concentration prediction models that incorporate attention mechanisms and spatial features.(4)This study proposes and constructs a combined PM2.5 concentration prediction model that integrates attention mechanism and spatiotemporal convolutional network,aiming to address the problem of the lack of consideration of the spatiotemporal correlations between the target station and nearby monitoring stations in the PM2.5 concentration prediction process.Firstly,the feature data of highly correlated monitoring stations are extracted and integrated using convolutional methods,obtaining the spatiotemporal feature information used for training and predicting.Then,the attention mechanism is added to the temporal convolutional network to enhance the important feature information and weaken the irrelevant feature information in the network model,thus improving the accuracy of PM2.5 concentration prediction for the target station.Based on the results of performance comparison experiments and ablation experiments,it can be concluded that this combined model has excellent data fitting ability in predicting PM2.5 concentration,not only effectively capturing nonlinear and non-stationary time series data but also being applicable for predicting PM2.5 concentration tasks over a relatively long period of time.