Long short-term memory - Fully connected (LSTM-FC) neural network for PM2.5 concentration prediction

Abstract

People have been suffering from air pollution for a decade in China, especially from PM2.5 (particulate matter with a diameter of less than 2.5 μm). Accurate prediction of air quality has great practical significance. In this paper, we propose a data-driven model, called as long short-term memory - fully connected (LSTM-FC) neural network, to predict PM2.5 contamination of a specific air quality monitoring station over 48 h using historical air quality data, meteorological data, weather forecast data, and the day of the week. Our predictive model consists of two components: (1) Using a long short-term memory (LSTM)-based temporal simulator to model the local variation of PM2.5 contamination and (2) Using a neural network-based spatial combinatory to capture spatial dependencies between the PM2.5 contamination of central station and that of neighbor stations. We evaluate our model on a dataset containing records of 36 air quality monitoring stations in Beijing from 2014/05/01 to 2015/04/30 and compare it with artificial neural network (ANN) and long short-term memory (LSTM) models on the same dataset. The results show that our LSTM-FC neural network model gives a better predictive performance.