Related Experiment Video
Updated: Jul 17, 2026
08:23
Analyzing the Photo-oxidation of 2-propanol at Indoor Air Level Concentrations Using Field Asymmetric Ion Mobility Spectrometry
Published on: June 14, 2018
8.9K
A novel approach for forecasting PM2.5 pollution in Delhi using CATALYST
Abhishek Verma1, Virender Ranga2, Dinesh Kumar Vishwakarma2
1Biometric Research Laboratory, Department of Information Technology, Delhi Technological University, Bawana Road, Delhi, -110042, India. abhishekcms08@gmail.com.
Environmental Monitoring and Assessment
|November 11, 2023
Summary
Accurate PM2.5 forecasting in Delhi is crucial for air quality management. A new hybrid model, CATALYST, combines CNN and Transformer architectures for improved prediction of fine particulate matter levels.
Area of Science:
- Environmental Science
- Data Science
- Artificial Intelligence
Background:
- Air pollution, specifically PM2.5, poses significant environmental and public health challenges in urban areas like Delhi.
- Accurate forecasting of PM2.5 concentrations is essential for developing effective pollution control strategies and informing policy decisions.
Purpose of the Study:
- To introduce a novel hybrid model, CATALYST (Convolutional and Transformer model for Air Quality Forecasting), for enhanced PM2.5 pollution prediction in Delhi.
- To leverage deep learning techniques, including Convolutional Neural Networks (CNN) and Transformer architectures, for time-series air quality forecasting.
- To address data challenges in environmental time-series analysis, such as missing or partial data, through an innovative input-output coupling approach.
Main Methods:
- A pre-trained CNN model is employed to extract visual features from PM2.5 time-series data.
- The CATALYST model, integrating CNN features with a Transformer architecture, is trained using a sliding window approach to capture temporal dependencies.
- The Transformer component utilizes positional encoding to analyze intricate patterns influenced by meteorological, geographical, and anthropogenic factors.
- A novel method for constructing input-output data pairs is implemented to handle incomplete environmental datasets.
Main Results:
- The CATALYST model demonstrated superior performance in forecasting PM2.5 pollution compared to traditional methods like ARIMA and LSTM.
- Experimental results indicate the model's effectiveness in capturing complex temporal dynamics within PM2.5 time-series data.
- The proposed data handling technique ensures comprehensive training datasets, even with missing or partial environmental information.
Conclusions:
- The CATALYST model presents a promising and accurate strategy for PM2.5 pollution forecasting in urban environments.
- This hybrid approach offers potential for real-world application in developing targeted pollution mitigation measures.
- Further validation with diverse real-world datasets is recommended to fully assess CATALYST's applicability and impact on environmental sustainability and public health.
