PaperlayerView exact revision on arXiv

Source-linked AI summary

Vehicle Energy Dataset (VED), A Large-scale Dataset for Vehicle Energy Consumption Research

G. S. Oh, David J. Leblanc, Huei Peng

arXiv:1905.02081v1physics.soc-ph

TL;DR

Public datasets lacked large-scale individual-vehicle trajectories paired with time-series fuel and energy data. VED releases de-identified real-world records from 383 vehicles and uses case studies to examine fuel-economy patterns and energy-saving opportunities.

  • Problem

    The authors identify a lack of publicly available large-scale datasets combining individual-vehicle trajectories with time-series fuel and energy data.

    Source: §I · p. 1

  • Method

    VED provides de-identified real-world trajectories and time-series vehicle energy data collected from 383 vehicles.

    Source: §I · p. 1, §VII · p. 10

  • Results

    Highway MPG averaged 2.92 times busy inner-city MPG, and −8°C increased fuel use 46% versus 20°C; near signalized intersections, ICEs used 287.5% more fuel than HEVs.

    Source: §VII · p. 10

  • Takeaways & Limitations

    The dataset supports data-driven energy-consumption models for eco-routing and eco-approach and departure algorithms, and offers research opportunities for CAV and autonomous-driving technologies.

    Source: §VII · p. 10

Abstract

from arXiv · show

We present Vehicle Energy Dataset (VED), a novel large-scale dataset of fuel and energy data collected from 383 personal cars in Ann Arbor, Michigan, USA. This open dataset captures GPS trajectories of vehicles along with their time-series data of fuel, energy, speed, and auxiliary power usage. A diverse fleet consisting of 264 gasoline vehicles, 92 HEVs, and 27 PHEV/EVs drove in real-world from Nov, 2017 to Nov, 2018, where the data were collected through onboard OBD-II loggers. Driving scenarios range from highways to traffic-dense downtown area in various driving conditions and seasons. In total, VED accumulates approximately 374,000 miles. We discuss participant privacy protection and develop a method to de-identify personally identifiable information while preserving the quality of the data. After the de-identification, we conducted case studies on the dataset to investigate the impacts of factors known to affect fuel economy and identify energy-saving opportunities that hybrid-electric vehicles and eco-driving techniques can provide. The case studies are supplemented with a number of examples to demonstrate how VED can be utilized for vehicle energy and behavior studies. Potential research opportunities include data-driven vehicle energy consumption modeling, driver behavior modeling, machine and deep learning, calibration of traffic simulators, optimal route choice modeling, prediction of human driver behaviors, and decision making of self-driving cars. We believe that VED can be an instrumental asset to the development of future automotive technologies. The dataset can be accessed at https://github.com/gsoh/VED.

I. INTRODUCTION

Public datasets lacked large-scale individual-vehicle trajectories paired with time-series fuel and energy data. VED addresses this gap with a year of real-world driving data collected from a diverse fleet.

I. INTRODUCTION

  • Public traffic datasets provide aggregate counts, while large-scale public datasets pairing individual-vehicle trajectories with time-series energy data were not available to the authors’ knowledge.Existing public datasets for machine vision, radar, and LiDAR did not supply these energy time series either.Source: §I · p. 1

II. VEHICLE ENERGY DATASET (VED)

  • VED publishes real-world fuel and energy consumption time series for diverse personal cars collected from November 2017 to November 2018.The dataset is presented as a resource for studying energy use across driving conditions.Source: §I · p. 1

A. Data Collection

  • OBD-II loggers collected GPS trajectories and vehicle signals from 383 cars over 373,964 miles, with ICE/HEVs accounting for 86.6% of mileage and PHEVs/BEVs 13.4%.The fleet included passenger cars and light trucks and drove from November 1, 2017, to November 9, 2018.Source: A. Data Collection · p. 2

B. Data Description

VED combines static vehicle information with dynamic, time-stamped driving records, while signal availability varies across vehicle groups.

B. Data Description

  • VED separates static vehicle parameters from dynamic time-stamped records, whose signals are categorized as GPS, standard OBD-II, and OEM-customized OBD-II.Tables II and III describe the dynamic and static data contents, respectively.Source: B. Data Description · p. 2, Table II · p. 3, Table III · p. 3
  • All vehicle groups report GPS, but ICE/HEVs lack battery and auxiliary-power signals, while EVs lack engine-related signals.The dynamic signal lists are divided into ICE/HEV, PHEV, and EV groups.Source: B. Data Description · p. 3
  • The fleet includes 264 ICE vehicles, 92 HEVs, 24 PHEVs, and 3 EVs, with distinct powertrain and charging characteristics.HEVs combine an engine and motor and rely on regenerative braking or engine power; PHEVs can also charge from wall power, while EVs operate solely on electricity.Source: B. Data Description · p. 3

C. Estimation of fuel consumption

Because OBD-II usually does not report fuel consumption directly, the paper estimates fuel-consumption rate from available engine signals and applies a fuel-trim correction.

C. Estimation of fuel consumption

  • Fuel-rate estimates use reported FuelRate when available, otherwise MAF, or MAF estimated from absolute load and engine speed when those signals are available.The algorithm’s inputs include FuelRate, MAF, absolute load, engine displacement, and engine RPM.Source: C. Estimation of fuel consumption · p. 4
  • The correction factor uses STFT and LTFT with AFR; when dynamic AFR is unavailable, the method uses 14.08 for E10 fuel.For vehicles with trim signals from two banks, their values are averaged; other noise-reduction methods may also be applied.Source: C. Estimation of fuel consumption · p. 4
  • For PHEVs and EVs, instantaneous electric energy consumption rate is obtained by multiplying battery voltage by current.Battery SOC and filtering can supplement the estimate.Source: C. Estimation of fuel consumption · p. 4

III. DE-IDENTIFICATION

GPS trip endpoints and vehicle details can expose private information. VED’s de-identification method aims to protect drivers while preserving useful spatial and temporal detail.

  • GPS coordinates at trip starts and ends can reveal participants’ homes, workplaces, time-specific locations, and points of interest.Source: §III · p. 4
  • Anonymization trades privacy against spatial-temporal resolution, which must remain sufficient to identify roads and routes from GPS traces.Source: §III · p. 4
  • The method suppresses or generalizes GPS trajectories and vehicle-specific details while preserving general powertrain information.Source: §III · p. 4

A. Random-fogging

VED de-identifies trip records by masking trip endpoints and restricting retained data geographically. These steps suppress potentially identifying locations while retaining routes within Ann Arbor.

A. Random-fogging

B. Geo-fencing

  • Geo-fencing retains recordings within a rectangular boundary around Ann Arbor, excluding rural routes where sparse roads and few unique paths could ease re-identification.Source: B. Geo-fencing · pp. 4–5
  • Fig. 4 depicts the geographic fence and trip bounds at the first and last major intersections, with GPS data outside those bounds removed.Source: Figure 4 · p. 5

C. Major Intersections Bounding

The section describes major-intersection trimming for privacy and examines fuel economy across driving environments and speeds. Its findings include higher highway MPG than on urban roads and lower MPG near intersections.

C. Major Intersections Bounding

IV. DRIVER BEHAVIORS ON FUEL CONSUMPTION

  • The driver-behavior analyses examine driving environments, speed, seasons, ambient temperature, and time of day in relation to fuel economy or travel speed.Source: §IV · p. 5

A. Influence of driving environments on Fuel Economy

B. Influence of Season and Temperature on Fuel Economy

VED data show seasonal and ambient-temperature patterns in fuel economy for ICE vehicles and HEVs. The observed seasonal difference is an overall fleet result; individual vehicle behavior can vary.

B. Influence of Season and Temperature on Fuel Economy

C. Influence of traffic density on MPG and speed

The analysis compares vehicle MPG and speed during rush-hour and free-flow time windows. Both measures differed between the two periods.

C. Influence of traffic density on MPG and speed

V. FUEL AND ENERGY USE OF HYBRID VEHICLES

VED case studies illustrate fuel-economy differences between ICE vehicles and HEVs and examine energy use during a PHEV trip. The examples include regenerative braking and distinct PHEV operating modes.

A. Fuel Economy: ICE Vehicle versus HEV

B. Case Study on PHEV

  • PHEVs typically operate in EV mode or Hybrid mode, and VED contains driving records from both modes.EV mode uses the electric motor alone; Hybrid mode uses both the engine and motor.Source: B. Case Study on PHEV · p. 8
  • HEVs did not output battery signals, so the paper directs researchers to PHEV and EV data for studying electric-vehicle energy consumption.Source: B. Case Study on PHEV · p. 8
  • 0.1265 kWh over 0.5192 miles corresponded to 138 MPGe for a PHEV trip conducted exclusively in EV mode, with no fuel consumed.The battery recharged during deceleration through regenerative braking and depleted during acceleration.Source: Figure 12 · p. 8, B. Case Study on PHEV · p. 9

VI. ECO-DRIVING

The eco-driving section presents eco-routing and eco-approach and departure as applications of VED. Its route case study illustrates how route choice relates to fuel economy and travel time.

VI. ECO-DRIVING

  • Eco-driving applications include eco-routing to select routes that minimize energy consumption and eco-approach and departure to optimize vehicle trajectories for energy use.Source: §VI · p. 9

A. Eco-routing

  • VED fuel and energy data can support consumption models by vehicle, trip, route, and link for energy-efficient, fastest, or shortest-route selection.Source: A. Eco-routing · p. 9
  • R1/R2 median MPG was 24.3/13.0, and median travel time was 130.0s/168.3s, illustrating route-dependent fuel and time outcomes.The compared trips used two routes with the same origin-destination pair.Source: Figure 14 · p. 9, Figure 13 · p. 9, A. Eco-routing · p. 9
  • VED captures diverse driving conditions and time-varying speed profiles, including intersection records relevant to eco-routing models and simulation-to-reality challenges.The paper notes that accurate energy-consumption modeling at intersections remains a challenge for eco-routing algorithms.Source: A. Eco-routing · p. 9

B. Eco-approach and Departure

VED combines de-identified vehicle trajectories and energy data with case studies that identify fuel-saving opportunities. The studies report energy-use differences across conditions and illustrate how signal timing and stops distinguish trips on the same route.

B. Eco-approach and Departure

  • EAD uses signal timing and phasing information communicated over V2X to seek energy-efficient trajectories through signalized intersections; prior work reported 40–50% average fuel-saving potential in free-flow scenarios.The cited prior work studied 609 real-world driving records.Source: B. Eco-approach and Departure · p. 9
  • In two same-route trips by a 2013 Honda Accord ICE, trip 1 consumed 32% less fuel than trip 2; trajectories suggest trip 2 stopped several times at signalized intersections, unlike trip 1.The respective fuel economies were 25.2 and 17.3 MPG; surrounding traffic was unobservable and may have affected the trips.Source: B. Eco-approach and Departure · pp. 9–10, Figure 15 · p. 10
  • The authors infer that catching the right traffic-signal phase and minimizing stops offer opportunities to save fuel and travel time, while noting that surrounding traffic was unobservable.The unobserved traffic may have affected the trips.Source: B. Eco-approach and Departure · p. 10

VII. CONCLUSION

  • VED contains second-by-second trajectories and energy data from 383 privately owned vehicles across diverse driving settings, with de-identification intended to protect personal information.The fleet includes ICE vehicles, HEVs, PHEVs, and EVs.Source: §VII · p. 10
  • Highway MPG averaged 2.92 times busy inner-city MPG; highest-MPG speed was 18 m/s; at −8°C, fuel use was 46% higher than at 20°C, and rush-hour trips used 67% more fuel for the same distance.These results describe variation by driving environment and speed, temperature, and time of day.Source: §VII · p. 10
  • ICE vehicles used 287.5% more fuel than HEVs near signalized intersections, and VED supports analysis of PHEV and EV battery and auxiliary power use.The case studies also identify routing and signalized-intersection fuel-saving opportunities that can be pursued using CAV technologies.Source: §VII · p. 10
  • VED can support data-driven energy models for diverse driving conditions and research on eco-routing and eco-approach and departure algorithms.The conclusion identifies these as research opportunities for CAV and autonomous-driving work.Source: §VII · p. 10
Loading 1905.02081v1…