Comparative Analysis of LSTM and GRU Models for Air Pollution Prediction in Shanghai

This paper presents a comparative analysis of two recurrent neural network (RNN) models, respectively with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), specifically focusing on predicting PM2.5/PM10 concentrations in Shanghai. Running PM2.5/PM10 prediction accurately and efficiently is important for public health and environmental management, while few research programs have constructed efficient forecasting models when facing a large amount of daily variation in air pollution. In this study, historical air quality data, meteorological factors, and other relevant environmental variables were concentrated to train and evaluate the models. After the construction of two different models, the performance of LSTM and GRU is compared based on various metrics, including Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared. The results indicate the strengths and limitations of each model in capturing the temporal dependencies in air quality data. Moreover, the study explores the impact of different input features and network configurations on prediction accuracy based on the real situation. The findings promote developing effective air quality forecasting tools and help local governments inform policy decisions aimed at mitigating air pollution in urban environments.