Abstract:Accurate prediction of particulate matter with aerodynamic diameter ≤ 2.5 μm (PM2.5) is important for environmental management and human health protection. In recent years, many efforts have been devoted to develop air quality predictions using the machine learning and deep learning techniques. In this study, we propose a deep learning model for short-term PM2.5 predictions. The salient feature of the proposed model is that the convolution in the model architecture is causal, where the output of a time step is only convolved with components of the same or earlier time step from the previous layer. The model also weighs the spatial correlation between multiple monitoring stations. Through temporal and spatial correlation analysis, relevant information is screened from the monitoring stations with a strong relationship with the target station. Information from the target and related sites is then taken as input and fed into the model. A case study is conducted in Nanjing, China from January 1, 2020 to December 31, 2020. Using historical air quality and meteorological data from nine monitoring stations, the model predicts PM2.5 concentrations for the next hour. The experimental results show that the predicted PM2.5 concentrations are consistent with observation, with correlation coefficient (R2) and Root Mean Squared Error (RMSE) of our model are 0.92 and 6.75 μg/m3. Additionally, to better understand the factors affecting PM2.5 levels in different seasons, a machine learning algorithm based on Principal Component Analysis (PCA) is used to analyze the correlations between PM2.5 and its influencing factors. By identifying the main factors affecting PM2.5 and optimizing the input of the predictive model, the application of PCA in the model further improves the prediction accuracy, with decrease of up to 17.2 % in RMSE and 38.6 % in mean absolute error (MAE). The deep learning model established in this study provide a valuable tool for air quality management and public health protection.

Fine particulate matter concentration prediction based on hybrid convolutional network with aggregated local and global spatiotemporal information: A case study in Beijing and Chongqing

Forecasting PM2.5 Using Hybrid Graph Convolution-Based Model Considering Dynamic Wind-Field to Offer the Benefit of Spatial Interpretability.

A Spatial-Temporal Causal Convolution Network Framework for Accurate and Fine-Grained PM2.5 Concentration Prediction

An adaptive adjacency matrix-based graph convolutional recurrent network for air quality prediction

Near-surface PM2.5 prediction combining the complex network characterization and graph convolution neural network

ST-CCN-PM2.5: fine-grained PM2.5 concentration prediction via spatial-temporal causal convolution network

Adaptive Scalable Spatio-Temporal Graph Convolutional Network for PM2.5 Prediction

ST-CCN-PM 2.5

Regional PM2.5 Prediction with Hybrid Directed Graph Neural Networks and Spatio-temporal Fusion of Meteorological Factors

A Hybrid CNN-LSTM Model for Predicting PM2.5 in Beijing Based on Spatiotemporal Correlation

Urban PM2.5 Concentration Prediction Via Attention-Based CNN–LSTM

Spatiotemporal causal convolutional network for forecasting hourly PM2.5 concentrations in Beijing, China

PM2.5 Concentration Prediction Based on CNN-BiLSTM and Attention Mechanism

72-hour real-time forecasting of ambient PM2.5 by hybrid graph deep neural network with aggregated neighborhood spatiotemporal information

A Hybrid Spatiotemporal Model Combining Graph Attention Network and Gated Recurrent Unit for Regional Composite Air Pollution Prediction and Collaborative Control

Spatio-temporal fusion of meteorological factors for multi-site PM2.5 prediction: A deep learning and time-variant graph approach

Spatiotemporal Adaptive Attention Graph Convolution Network for City-Level Air Quality Prediction.

The Air Quality Prediction Based on a Convolutional LSTM Network.

A Spatiotemporal Recurrent Neural Network for Prediction of Atmospheric PM2.5: A Case Study of Beijing

Air Pollution Prediction Via Graph Attention Network and Gated Recurrent Unit

Improved prediction of hourly PM2.5 concentrations with a long short-term memory and spatio-temporal causal convolutional network deep learning model