Low-Cost Model for the Estimation of Pollutant Emissions Based on GPS and Machine Learning

Rivera-Campoverde, Néstor; Sanz, José Muñoz; Arenas-Ramirez, Blanca · 2023 · Crossref

DOI: 10.1007/978-3-031-38563-6_27

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of accurately estimating pollutant emissions from internal combustion engine vehicles under real driving conditions without relying on expensive, extensive measurement campaigns or long-term vehicle instrumentation. Current regulations require Real Driving Emissions (RDE) testing because laboratory tests often underestimate emissions by ignoring factors like traffic patterns, route selection, and driving behavior. The authors propose a low-cost model that utilizes Global Positioning System (GPS) data and machine learning techniques to estimate emissions, aiming to provide a robust alternative to traditional methods that depend on On-Board Diagnostics (OBD) or complex portable emission measurement systems. The methodology involves a two-phase experimental design conducted in Cuenca, Ecuador, using a 2018 Kia Sportage. First, two RDE-compliant routes comprising urban, rural, and highway segments were driven while collecting data via a portable emission measurement system (PEMS), an OBD recorder, and a GPS logger. Data from the first route was used to train the model, while the second route served for validation. Since GPS data lacks direct engine parameters like throttle position or gear selection, the authors developed a preprocessing pipeline. They used the K-means algorithm to cluster vehicle speed and engine RPM ratios to label gear states, followed by a classification tree to predict the driver’s selected gear based on GPS-derived acceleration and speed. These predicted gears and other GPS-derived variables were then used to train four Artificial Neural Networks (ANNs) to estimate emissions for CO2, CO, HC, and NOx. The results demonstrate that the machine learning model effectively estimates emissions with high accuracy. The ANNs achieved R² determination indices of 0.935 for CO2, 0.923 for CO, 0.914 for HC, and 0.943 for NOx. When validated against RDE measurements, the model produced relative errors of 0.0976% for CO2, −0.2187% for CO, 0.2249% for HC, and −0.1379% for NOx. The model was further applied to a dataset of 1,218 km of random driving. Comparisons with the International Vehicle Emissions (IVE) model and direct RDE measurements showed that the GPS-based model yielded emission factors closer to real-world conditions, particularly in low-speed urban environments where the IVE model tended to overestimate emissions. The significance of this work lies in its provision of a cost-effective, scalable method for estimating vehicular emissions using widely available GPS data. By successfully deriving complex engine states from simple positioning signals, the model reduces the barrier to entry for emission monitoring. The authors conclude that this approach is more robust to varying traffic conditions than existing models like IVE and is suitable for calculating emission factors and constructing vehicular emission inventories, offering a practical tool for urban air quality management and regulatory compliance assessment.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement