In order to join virtual sessions, you must be registered and logged-in(Were you registered for the in-person meeting in Denver? if yes, just log in.) 
Note: All session times are in Mountain Daylight Time.

Predicting hourly ground-level PM2.5 concentrations using a spatiotemporal attention-based recurrent neural network

Authors: Luwei Feng*, Wuhan University, Qingyun Du, Wuhan University
Topics: Environment, China
Keywords: PM2.5 prediction, attention mechanism, recurrent neural network
Session Type: Poster
Presentation File: No File Uploaded


Lone-term exposure to high concentrations of particulate matter with a diameter < 2.5 μm (PM2.5) have a severe effect on public health, thereby increasing risks of respiratory disease, cardiovascular disease and reduced lung function. The accurate prediction of PM2.5 is significant in understanding PM2.5 changing tendency, advising people on their travels and helping control regional air pollution. In this study, we predict the PM2.5 concentrations over several future hours based on sequential data, hourly PM2.5 data and meteorological data, and non-sequential data, such as points of interest (POIs), structure of road networks and population distribution. A data-driven model is proposed by integrating spatial and temporal attention layers into recurrent neural network (RNN). Spatial attention layer considers the complex spatial correlation of data at the same time while temporal attention layer captures the continuously changing characteristics of sequential data. The site-based PM2.5 data and other predictive data about Beijing, the capital of China, from Jan. 1, 2018 to Dec. 31, 2018 are used to verify the validity of the proposed model. We also compare the performance of the model with ANN and long short-term memory (LSTM). The results show that our model gives a better predictive performance.

To access contact information login