[Paper Review] Deep Air Quality Forecasting Using Hybrid Deep Learning Framework

Shengdong Du, Tianrui Li+2 more|arXiv (Cornell University)|Dec 12, 2018

Air Quality Monitoring and Forecasting35 references47 citations

TL;DR

This paper introduces a hybrid deep learning model that combines 1D CNNs and Bi-LSTM networks to forecast PM2.5 by learning local spatial-temporal features and interdependencies in multivariate air quality time series.

ABSTRACT

Air quality forecasting has been regarded as the key problem of air pollution early warning and control management. In this paper, we propose a novel deep learning model for air quality (mainly PM2.5) forecasting, which learns the spatial-temporal correlation features and interdependence of multivariate air quality related time series data by hybrid deep learning architecture. Due to the nonlinear and dynamic characteristics of multivariate air quality time series data, the base modules of our model include one-dimensional Convolutional Neural Networks (1D-CNNs) and Bi-directional Long Short-term Memory networks (Bi-LSTM). The former is to extract the local trend features and spatial correlation features, and the latter is to learn spatial-temporal dependencies. Then we design a jointly hybrid deep learning framework based on one-dimensional CNNs and Bi-LSTM for shared representation features learning of multivariate air quality related time series data. We conduct extensive experimental evaluations using two real-world datasets, and the results show that our model is capable of dealing with PM2.5 air pollution forecasting with satisfied accuracy.

Motivation & Objective

• Motivate accurate forecasting of air quality (PM2.5) for early warning and control.
• Develop a model that captures spatial-temporal correlations in multivariate air quality time series.
• Integrate 1D CNNs and Bi-LSTM into a unified framework for shared representation learning.

Proposed method

• Use one-dimensional CNNs to extract local trend features and spatial correlations.
• Use Bi-directional LSTM networks to learn spatial-temporal dependencies.
• Design a jointly hybrid framework that combines CNN and Bi-LSTM for shared representations of multivariate time series.
• Evaluate the model on two real-world datasets with PM2.5 forecasting tasks.

Experimental results

Research questions

• RQ1 Can a hybrid CNN-Bi-LSTM framework effectively capture local and global spatial-temporal dependencies in multivariate air quality data?
• RQ2 Does the proposed architecture improve PM2.5 forecast accuracy on real-world datasets compared to baseline models?

Key findings

• The model demonstrates satisfactory accuracy in PM2.5 forecasting on real-world data.
• The architecture learns both local trend features and interdependencies among multivariate air quality time series.
• Experimental evaluation indicates effectiveness of the hybrid framework for air quality forecasting.