Spatio-Temporal Forecasting of $$PM_{2.5}$$ via Spatial-Diffusion Guided Encoder-Decoder Architecture

This repository contains the code for the research paper "Spatio-Temporal Forecasting of $$PM_{2.5}$$ via Spatial-Diffusion guided Encoder-Decoder Architecture" accepted for presentation at "CODS-COMAD'24".

Table of Contents

• Abstract
• Dataset
• Requirements
• Run
• Citation

Abstract

In many problem settings that require spatio-temporal forecasting, the values in the time-series not only exhibit spatio-temporal correlations but are also influenced by spatial diffusion across locations. One such example is forecasting the concentration of fine particulate matter ($$PM_{2.5}$$) in the atmosphere which is influenced by many complex factors, the most important ones being diffusion due to meteorological factors as well as transport across vast distances over a period of time. We present a novel Spatio-Temporal Graph Neural Network architecture, that specifically captures these dependencies to forecast the $$PM_{2.5}$$ concentration. Our model is based on an encoder-decoder architecture where the encoder and decoder parts leverage gated recurrent units (GRU) augmented with a graph neural network (TransformerConv) to account for spatial diffusion. Our model can also be seen as a generalization of various existing models for time-series or spatio-temporal forecasting. We demonstrate the model's effectiveness on two real-world $$PM_{2.5}$$ datasets: (1) data collected by us using a recently deployed network of low-cost $$PM_{2.5}$$ sensors from 511 locations spanning the entirety of the Indian state of Bihar over a period of one year, and (2) another publicly available dataset that covers severely polluted regions from China for a period of 4 years. Our experimental results show our model's impressive ability to account for both spatial as well as temporal dependencies precisely. The code is publicly available at Github Repository.

Dataset

This study is based on two similar real world datasets:

• Bihar, India: A dataset collected using 511 low-cost sensors deployed across the entire region of Bihar for a year (01/05/2023 - 30/04/2024).
• KnowAir Dataset: Dataset covering severly polluted regions in China, spanning across 4 years (01/01/2015 - 31/12/2018). Dataset is available at KnowAir.

Requirements

Prerequisites

• Python 3.10.13 or above
• GPU recommended for training (trained on NVIDIA A30 with 24G memory)

Installation

• Clone this repository

  git clone https://github.com/malayp717/pm2.5.git
  cd pm2.5
  

• Install dependencies:

  pip install -r requirements.txt
  

Experimental Setup

• Choose the model you want to run, and make required changes in either bihar_config.yaml or china_config.yaml, depending on your location preference.

  train:
      model: GRU
      model: GC_GRU
      model: GraphConv_GRU
      model: GNN_GRU
      model: Attn_GNN_GRU
  

Run

Training

python -u train.py --config <config_file>.yaml

Make sure to choose the correct configuration file:

• bihar_config.yaml: Use this file for training on the Bihar dataset.
• china_config.yaml: Use this file for training on the China dataset.

Evaluation

python -u stats.py --config <config_file>.yaml

Citation

@misc{pandey2024spatiotemporalforecastingpm25spatialdiffusion,
      title={Spatio-Temporal Forecasting of PM2.5 via Spatial-Diffusion guided Encoder-Decoder Architecture}, 
      author={Malay Pandey and Vaishali Jain and Nimit Godhani and Sachchida Nand Tripathi and Piyush Rai},
      year={2024},
      eprint={2412.13935},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2412.13935}, 
}