Estimation of Pollutant Emissions in Real Driving Conditions Based on Data from OBD and Machine Learning

Area of Science:

• Environmental Science
• Automotive Engineering
• Data Science

Background:

• Current models for estimating vehicle emissions often require extensive and costly measurement campaigns.
• There is a need for accurate emission estimation methodologies applicable to real-world driving conditions.

Purpose of the Study:

• To develop and validate a novel methodology for estimating vehicle emissions in real driving conditions.
• To leverage onboard diagnostics data and machine learning for pollutant estimation.
• To reduce reliance on large-scale measurement campaigns for emission data.

Main Methods:

• Utilized a data logger for driving data and a portable emissions measurement system for emissions data.
• Employed artificial neural networks trained on collected data for emission estimation.
• Applied random forest techniques to determine variable importance and K-means clustering for gear selection classification.
• Developed a classification tree model incorporating driver gear selection.

Main Results:

• The developed machine learning models were trained on a dataset covering 1218.19 km of driving.
• Results demonstrated comparable accuracy to the International Vehicle Emissions (IVE) model and real-world tests.
• The model showed robustness across diverse traffic conditions, particularly effective at lower average driving speeds typical of urban environments.

Conclusions:

• The proposed methodology provides a reliable approach for estimating vehicle emissions under real driving conditions.
• This method offers a cost-effective alternative to traditional measurement campaigns.
• The findings have potential applications in vehicle homologation and the creation of vehicular emission inventories.