LSTM-TCN-Attention Air Quality Prediction Model Based On Factor Analysis

With the acceleration of China’s industrialization process,air pollution has become increasingly severe,greatly affecting people’s quality of life.The concentration of PM2.5 in the air is an important monitoring indicator for measuring urban air quality.Establishing a reliable air quality prediction model and predicting changes in pollutant concentration in advance is of great significance for effectively addressing air pollution issues.However,existing deep learning models still have problems such as poor fitting accuracy and inability to focus on important information.Based on the above background information and the current problems in air quality prediction,this thesis mainly focuses on the following two tasks:The first task is to propose an air quality prediction model based on Long Short-Term Memory(LSTM)and Temporal Convolutional Network(TCN).In response to the air quality prediction task in Sanmenxia City,this thesis combines LSTM and TCN models in parallel to form a comprehensive LSTMTCN model.Through this comprehensive model,the long-term dependent memory ability of LSTM model and the efficient local pattern capture ability of TCN model are fully utilized,and the deep rules among captured data are used to predict the future PM2.5 concentration in Sanmenxia City.By comparing with LSTM and TCN models,the results show that the goodness of fit and prediction effect of the combined LSTM-TCN model are better than those of the single model.The second task is to propose an LSTM-TCN-Attention air quality prediction model based on factor analysis.The LSTM TCN parallel model did not consider the importance of each element in the sequence.The attention mechanism model based on factor analysis fills this gap.By calculating the importance of each element in the sequence through factor analysis,the model can focus more on those elements that contribute more to the prediction results,thereby improving its prediction performance.By implementing a fixed weight attention mechanism,effective information in input data can be better utilized,key information can be focused,and the predictive performance of the model can be improved.The goodness of fit,root mean square error,average absolute error,and other evaluation indicators of the LSTM-TCNAttention model based on factor analysis and the previous three models are compared through example verification,which proves the feasibility and effectiveness of the model.Moreover,the model has higher calculation efficiency and stronger ability to extract important features,which can further improve the overall accuracy of air quality prediction.