[Paper Review] Extended vehicle energy dataset (eVED): an enhanced large-scale dataset for deep learning on vehicle trip energy consumption

Shiliang Zhang, Dyako Fatih+3 more|arXiv (Cornell University)|Mar 16, 2022

Traffic Prediction and Management Techniques4 citations

TL;DR

This paper introduces the extended Vehicle Energy Dataset (eVED), a large-scale, GPS-accurate, and feature-rich dataset derived from the original VED by enhancing raw vehicle trip records with external geospatial attributes such as speed limits, elevation, intersections, and traffic infrastructure via map matching and API integration. Using this enriched data, the authors train an LSTM-based deep learning model that achieves a median speed prediction RMSE of 8.9078 on test segments, demonstrating the dataset’s utility for energy consumption modeling and privacy-preserving traffic behavior analysis.

ABSTRACT

This work presents an extended version of the Vehicle Energy Dataset (VED), which is a openly released large-scale dataset for vehicle energy consumption analysis. Compared with its original version, the extended VED (eVED) dataset is enhanced with accurate vehicle trip GPS coordinates, serving as a basis to associate the VED trip records with external information, e.g., road speed limit and intersections, from accessible map services to accumulate attributes that is essential in analyzing vehicle energy consumption. In particularly, we calibrate all the GPS trace records in the original VED data, upon which we associated the VED data with attributes extracted from the Geographic Information System (QGIS), the Overpass API, the Open Street Map API, and Google Maps API. The associated attributes include 12,609,170 records of road elevation, 12,203,044 of speed limit, 12,281,719 of speed limit with direction (in case the road is bi-directional), 584,551 of intersections, 429,638 of bus stop, 312,196 of crossings, 195,856 of traffic signals, 29,397 of stop signs, 5,848 of turning loops, 4,053 of railway crossings (level crossing), 3,554 of turning circles, and 2,938 of motorway junctions. With the accurate GPS coordinates and enriched features of the vehicle trip record, the obtained eVED dataset can provide a precise and abundant medium to feed a learning engine, especially a deep learning engine that is more demanding on data sufficiency and richness. Moreover, our software work for data calibration and enrichment can be reused to generate further vehicle trip datasets for specific user cases, contributing to deep insights into vehicle behaviors and traffic dynamics analyses. We anticipate that the eVED dataset and our data enrichment software can serve the academic and industrial automotive section as apparatus in developing future technologies.

Motivation & Objective

• To address the lack of large-scale, open, and richly annotated datasets for vehicle energy consumption research.
• To improve the accuracy and utility of the original Vehicle Energy Dataset (VED) by correcting noisy GPS traces and enriching them with external geospatial attributes.
• To enable deep learning applications in transportation research by providing a data-rich, privacy-conscious dataset with precise spatiotemporal features.
• To demonstrate the practical value of eVED through a deep learning model that estimates vehicle speed using only publicly available information.
• To release both the enhanced dataset and open-source software for future data enrichment and reuse in academic and industrial research.

Proposed method

• Applying Hidden Markov Model (HMM)-based map matching to correct noisy GPS coordinates in the original VED dataset, aligning them to the most likely road segments.
• Using QGIS, OpenStreetMap (OSM), Overpass API, and Google Maps Elevation API to extract 12+ million geospatial attributes, including speed limits, elevation, intersections, and traffic infrastructure.
• Integrating external features such as road gradient, traffic signals, stop signs, bus stops, and crossings into the VED records through spatial querying and geospatial processing.
• Calibrating and enriching the dataset using a modular software pipeline that supports reproducibility and reuse for custom vehicle datasets.
• Training a long short-term memory (LSTM) network on 7,267 training and 951 testing trip segments, each with 240 consecutive feature records, to estimate median vehicle speed.
• Configuring the LSTM model with 5,000 hidden units, 8 input features (including time of day, speed limit, gradient, and proximity to traffic points), and ReLU/tanh activation functions for optimal performance.

Experimental results

Research questions

• RQ1 Can a large-scale vehicle energy dataset be significantly enhanced with accurate GPS and rich geospatial attributes to support deep learning applications?
• RQ2 How effectively can publicly available geospatial data be integrated with raw vehicle trip records to enrich feature sets for traffic and energy modeling?
• RQ3 To what extent can a deep learning model trained on publicly available features accurately estimate vehicle speed without relying on sensitive internal vehicle data?
• RQ4 What is the predictive performance of a deep learning model on vehicle speed estimation when trained exclusively on open, privacy-safe features?
• RQ5 Can the data enrichment pipeline be reused to generate tailored datasets for specific research or industrial use cases?

Key findings

• The eVED dataset contains 12,609,170 records of road elevation, 12,203,044 of speed limit, and 12,281,719 of directional speed limit, significantly enriching the original VED.
• The dataset includes 584,551 intersections, 429,638 bus stops, 195,856 traffic signals, and 29,397 stop signs, enabling detailed traffic behavior analysis.
• The LSTM model trained on eVED achieved a root mean squared error (RMSE) of 8.9078 in predicting median vehicle speed over 4-minute trip segments on the test set.
• The model’s performance demonstrates that accurate speed estimation is feasible using only publicly available geospatial and temporal features, minimizing privacy risks.
• The software pipeline for GPS calibration and feature enrichment is open-sourced and reusable for generating custom datasets with similar richness.
• The eVED dataset enables advanced research in energy-efficient route planning, vehicle behavior modeling, and traffic dynamics, particularly for deep learning applications.