Vehicle-specific power-based emission modeling using unsupervised machine learning techniques

Precise evaluation of vehicular emissions in real-world conditions is essential for assessing air quality and determining the efficacy of emission control policies. A critical research gap exists regarding the comparative study between Bharat Stage VI (BS-VI) and BS-VI compliant vehicles in various diverse Indian driving conditions. Laboratory-based measurement often fails to captures detailed emission profiles in complex traffic conditions. This study addresses this gap by analyzing the emissions from BS-IV petrol, BS-IV diesel, and BS-VI petrol vehicles in Indian driving conditions using a portable emission measurement system (PEMS). We developed a novel framework that integrates vehicle specific power (VSP) with unsupervised techniques to identify and analyze the distinct emission profiles over various driving conditions. The best clustering algorithm's results (k means) were used to compare and assess the emissions characteristics of BS-VI and BS-IV vehicles under various driving conditions. Result showed that CO₂ and NO_x emissions were highest for all three vehicle types during transitions from idle to minor acceleration and lowest during idling/creeping. BS-VI petrol vehicles demonstrated a substantial decrease (25 to 90%) in NO_x emissions compared to BS-IV petrol vehicles across different driving conditions. However, the difference in CO₂ emissions between BS-IV and BS-VI petrol vehicles was minimal (5%). The developed models can be a decision-support tool for policymakers to encourage responsible driving behavior and reduce vehicular emissions. This research is notable for its accurate, data-driven representation of real-world emissions. It offers policymakers a crucial tool for assessing emission regulations and driving patterns, encouraging eco-friendly driving practices.