Assessing energy consumption impacts of traffic shifts based on real-world driving data

Transportation Research Part D 62 (2018) 489–507

Contents lists available at ScienceDirect

Transportation Research Part D journal homepage: www.elsevier.com/locate/trd

Assessing energy consumption impacts of traffic shifts based on real-world driving data

T

⁎

Marta Fariaa, , Catarina Rolima,b, Gonçalo Duartea,b, Tiago Fariasa, Patrícia Baptistaa,b a

LAETA, IDMEC, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1, 1049-001 Lisboa, Portugal IN+, Center for Innovation, Technology and Policy Research, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1, 1049-001 Lisboa, Portugal b

A R T IC LE I N F O

ABS TRA CT

Keywords: On-board vehicle monitoring Vehicle dynamics Traffic volumes Energy consumption Scenarios Traffic shifts

Information and communication technologies used for on-board vehicle monitoring have been adopted as an additional tool to characterize mobility flows. Furthermore, traffic volumes are traditionally measured to understand cities traffic dynamics. This paper presents an innovative methodology that uses an extensive and complementary real-world dataset to make a scenariobased analysis allowing assessing energy consumption impacts of shifting traffic from peak to offpeak hours. In the specific case of the city of Lisbon, a sample of 40 drivers was monitored for a period of six months. The obtained data allowed testing the impacts of increasing the percentage of traffic shifting from peak to off-peak hours in energy consumption. Both average speed and energy consumption variations were quantified for each of the tested percentages, allowing concluding that for traffic shifts of up to 30% a positive impact in consumption can be observed. In terms of potential gains associated to shifting traffic from peak hours, reductions in energy consumption from 0.1% to 0.4% can be obtained for traffic volumes shifts from 5 to 30%. Overall, the maximum reduction in energy consumption is achieved for a 20% traffic shift. Average speed variation follows the same trend as energy consumption, but in the opposite direction, i.e. instead of decreasing, average speed increases. For the best case scenario, considering only the sections of roads with traffic sensors, a 1.4% reduction in trip time may be achieved, as well as savings of up to 6 l of fuel and 14.5 kg of avoided CO2 emissions per day.

1. Introduction The transportation sector is responsible for 40% of final energy consumption in Portugal, with the road transportation sector being responsible in 2013 for 81% of that energy consumption (European Commission, 2015). This is mainly due to the transportation sector dependency on fossil fuels, with inherent consequences regarding disruptions of energy supply and price volatility. Thus, there is a strong drive to reverse this trend, and the most traditional approach relies on promoting alternative vehicle technologies and energy pathways. However, focusing on a better use of the vehicle, in order to educate driver towards a more sustainable usage of the transport system, might provide additional opportunities for energy savings and avoided emissions. In this sense, information and communication technologies (ICT) can potentially be a powerful tool to promote change in the transportation sector, as already presented in previous studies (Geenhuizen, 2009; Kompfner et al., 2008). ICT applied to the transportation sector can be differentiated as fulfilling different tasks (Geenhuizen, 2009): supporting choices of vehicle drivers and/or passengers (for instance, on-

⁎ Corresponding author at: IDMEC/IST, Dept. of Mechanical Engineering, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, Pav. De Mecânica I, 2° Andar, 1049-001 Lisbon, Portugal. E-mail address: [email protected] (M. Faria).

https://doi.org/10.1016/j.trd.2018.03.008

1361-9209/ © 2018 Elsevier Ltd. All rights reserved.

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

road information or routing advice regarding traffic, driving advice to reduce fuel use); reducing options or limiting drivers’ behavior (e.g. avoiding certain parts of the road network, or limiting driving speed); alerting drivers and/or passengers without restraining their behavior (e.g. several modes of advanced driver assistance); and serving to take over drivers’ decisions, either fully or partly (such as electronic bonding of cars or in intelligent speed adaptation). At a more basic level ICT can be used to unobtrusively monitor driving behavior and mobility patterns, thus helping to develop and calibrate traffic, vehicle or behavioral models using real world data. One possibility of ICT use are on-board vehicle user aid devices, which can be used, on a first stage, to characterize driving patterns and, on a second stage, to educate the driver to improve energy efficiency and reduce costs and environmental impacts associated to urban mobility. On the driving patterns characterization, several studies have used real-world driving data to find representative driving cycles (André et al., 1995; Della Ragione and Meccariello, 2017; Gonder et al., 2007; Pelkmans and Debal, 2006; Sevtsuk and Ratti, 2010; Tzirakis et al., 2006; Wang et al., 2012; Wang et al., 2008). Wang et al. (2012) used large-scale mobile phone data, with detailed Geographic Information System (GIS) data, to assess road usage patterns in urban areas. The authors found that the major usage of each road segment can be traced to its own driver sources. Based on this finding, the authors proposed a network of road usage by defining a bipartite network framework. This novel framework allowed creating a strategy that achieves a significant reduction of the travel time across the entire road system, compared to a benchmark approach. Sevtsuk and Ratti (2010) also used large-scale mobile phone data to confirm that there is significant regularity in urban mobility at different hours, days, and weeks. The authors focused on the longitudinal activity patterns of network cells rather than individual users to conclude that besides the differences between weekday and weekend travels, hourly, daily, and weekly activity distribution patterns in a city can also be significantly distinguished. Furthermore, the authors found that different hours of the day affect the activity levels of diverse areas of the city differently. Other studies have focused on how driving patterns influence emissions and fuel consumption (Ericsson, 2001; Jensen, 1995; Smidfelt-Rosqvist, 2003; Zhang et al., 2014). Traffic research has also studied the effect of traffic planning and street design on driving patterns (Hallmark et al., 2002; Smidfelt-Rosqvist, 2003). For example, Brundell-Freij and Ericsson (2005) used a data set of over 14,000 driving patterns registered in actual traffic to assess the influence of street characteristics, driver category and vehicle performance on urban driving patterns and, consequently, on emissions and fuel consumption. Several parameters were collected namely, vehicle speed, engine speed, ambient temperature and location via GPS. The GPS information allowed assigning detailed street and traffic attributes to the driving patterns. Also information concerning the driver and the car were included in the database. With all this information the authors were able to conclude that street and traffic environment affect driving behavior in connection with driver variables (such as age and gender) and car performance. For instance, the authors found that higher density of junctions controlled by traffic lights have a decreasing effect on average speed (decreases by 16 km/h with higher density). Also the street function was found to have effects on average speed, with speed being lower on local streets than on arterials. When considering the percentage of time at acceleration levels greater than 1.5 m/s2, which considerably impacts fuel consumption and emissions, the largest effect was found for the density of junctions controlled by traffic lights. The higher the density, the greater the percentage (4% increase) of high accelerations. This parameter was also found to be lower (1.8%) at a speed limit of 90 km/h. Other studies have focused on the vehicle, the driver and/or the traffic environment in order to assess the relation between driving patterns and emissions and fuel consumption (André et al., 1995; Brundell-Freij and Ericsson, 2005; de Vlieger et al., 2000; Fontaras et al., 2017; Tzirakis et al., 2007). One of the few studies assessing the effect of traffic conditions on fuel consumption is the one by de Vlieger et al. (2000). The authors considered the effect of driving behavior along with traffic conditions to assess their influence on fuel consumption and emissions, concluding that traffic condition has a major effect on fuel consumption and emissions. For intense traffic conditions an increase of up to 45% on fuel consumption was observed. However, for this study a small test fleet was used and only two traffic conditions were considered (rush hours and Sundays) to conclude on the fuel consumption increase. In this sense, there is a lack of studies using extensive and complementary real-world monitoring databases to establish correlations between traffic volumes, street characteristics and vehicle dynamics. Considering this, the aim of this paper was to assess the energy consumption impacts of shifting traffic from peak hours to off-peak hours, by connecting traffic and vehicle dynamic data in an innovative approach, using an extensive real-world database for the city of Lisbon, Portugal. Such correlations allowed testing the implementation of scenarios regarding the reduction in traffic volumes during rush hours, by also assessing the impacts in energy consumption. 2. Methods and data The methods applied and data collected for this work are described in the following sections, namely through a detailed characterization of the monitored drivers sample, of the data collection and of the performed data analysis. A generic overview of the data collected and variables considered is presented in Fig. 1. Fig. 2 illustrates the data analysis process, explaining how the collected data was used to create profiles that were modified to account for differences in traffic volumes between hours. Full lines refer to raw data or main calculations while dashed lines refer to intermediate calculations. 2.1. Data collection In order to obtain real-world driving data an on-board diagnostics (OBD) system connected to the vehicle OBD port was used. The tool used was the i2d (intelligence to drive) device, which is a non-intrusive system developed in an R&D project supported by the 490

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Data source

Variables

Google Maps Geocoding API

Street and city

Real world driving data

On-board data acquisition I2D device

Coordinates (lat/long), speed, acceleration, slope, VSP, etc.

Infrastructure characterization

Municipality

Hierarchical street level and traffic volumes

Fig. 1. Generic overview of data and variables.

Data analysis

Case study characterization Average hourly traffic volume [veh/h] Hourly traffic volume per traffic sensor [veh/h] Max traffic volume per traffic sensor [veh/h]

Scenarios – traffic shifting

Hourly v/c ratio

Avg speed with idling [km/h] Avg speed without idling [km/h]

Speed Delta

VSP [W/kg]

Fuel consumption [l/100km]

Avg speed [km/h]

Probability of ocurrence of v/c ratios Probability of having congested conditions

NEW average speed [km/h]

Fuel consumption vs average speed

NEW fuel consumption [l/100km]

NEW traffic volume [veh/h]

Impacts in energy consumption of shifting traffic

Fig. 2. Overview of data analysis process.

Portuguese Innovation Support Fund for Energy Efficiency (FAI) (see Fig. 3) (iTds and IDMEC-IST, 2011). The i2d device collects, measures and automatically transmits with a 1 Hz frequency micro-scale driving data from the vehicle, including driving dynamics (speed, acceleration) and engine data (mass or air flow, engine rpm and load, throttle position, etc.). Additionally to the OBD connection, this device has dedicated sensors that allow collecting data on location (GPS) and road topography (barometric altimeter). The variation of the altitude is collected with an accuracy of under 1 m and georeferenced information has an accuracy of about 2.5 m (LiveDrive, 2015). For the purpose of this work, the information considered included speed and acceleration collected from the OBD, as well as road grade and location determined though a barometric altimeter and GPS, respectively. The present study was publicized through a pamphlet that presented the general idea of the project and its objectives, asking for volunteers. From the drivers that volunteered for the project, 40 were selected to participate. Taking into account the representativeness of the sample, drivers were selected in order to have a homogenous sample in terms of age (covering most age groups – from 25 years to 70 years), gender and driving experience, as presented in Table 1. This sample of 40 drivers was monitored for a period of six months (between 2014 and 2015) in the Metropolitan area of Lisbon, Portugal, using the i2d device. Throughout the monitoring period, the drivers used their own vehicles. Regarding the monitored vehicles, most were diesel (65%) with < 10 years (Euro 4 or newer) (as presented in Table 2). Regarding the vehicles type, it ranged from small city cars to sport utility vehicles (SUVs), with family cars (40% of the vehicles) being the most common. These vehicles were used only as probe vehicles, allowing characterizing the circulation conditions in the city in detail. In the complete monitoring period, a total of 5352 days were monitored (data for all drivers), during which over 25,000 trips were made, travelling a total of approximately 285,000 km in 7150 h. To characterize the road infrastructure, GPS data (coordinates and vehicle course) was used to determine what road and direction the vehicle was driving in. In order to associate each pair of coordinates on a second-by-second basis to a street name, a software was

Fig. 3. On-board monitoring i2d device used. 491

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Table 1 Characterization of the monitored drivers. Gender (N)

Age (years)

Driving experience (years)

Female

Male

Mean

Std. Deviation

Mean

Std. Deviation

18

22

43.4

12.2

23.7

11.8

Table 2 Characterization of the monitored vehicles. Number of vehicles Euro standard

Euro Euro Euro Euro

3 4 5 6

7 18 14 1

Fuel type

Diesel Gasoline Gasoline/LPG

26 13 1

developed in Visual Basic for Applications (VBA), within Microsoft Excel, by using the Google Maps Geocoding API (Google, 2016). The resulting information allowed establishing a connection between vehicle dynamics and the specific road where the vehicles circulated. The next step was to classify the type of road. Based on the GPS coordinates, a road hierarchical street level (Câmara Municipal de Lisboa, 2005) was attributed to each second of driving data. Four hierarchical street levels that define street function were considered: local streets (level 4), distributor and collector streets (level 3), minor arterial streets (level 2) and arterial streets (level 1) (Federal Highway Administration, 2013). Local streets offer the lowest level of mobility and the highest level of land access service. Distributor and collector streets collect traffic from local streets and channels it into the arterial system providing both land access service and traffic circulation. Minor arterial streets serve trips of moderate length at a somewhat lower level of travel mobility than principal arterials providing intracommunity continuity (Federal Highway Administration, 2013). Finally, level 1 streets (arterials) are comparable to highways with less variable circulation conditions than fully urban streets. While level 1 streets typically have freeflow conditions with little influence from the infrastructure and driving context on the driving conditions, urban streets have specific characteristics (such as high number of intersections – both signalized and non-signalized –, interaction with pedestrians, lower maximum allowed driving speed, etc.) which highly influence the vehicle dynamics. The match between hierarchical street levels to each street was used as defined for the city of Lisbon by the Municipality (Câmara Municipal de Lisboa, 2005). Another very relevant data source to further characterize the circulation patterns was related to traffic volumes. The Municipality of Lisbon provided traffic volumes data in a 15-min basis (Source: CML, Traffic and Urban Mobility Department), measured by intrusive traffic sensors (automatic traffic monitoring systems) installed on specific streets which are representative of typical traffic conditions within Lisbon. To quantify the average daily traffic (ADT), data from 84 traffic sensors in 49 locations was used. Traffic volumes data for level 1 streets were not available; therefore, the data considered relates only to level 2, 3 and 4 streets. The majority of the traffic sensors are installed in level 2 and level 3 streets (43 and 36 traffic sensors, respectively), while in level 4 streets only 5 traffic sensors are implemented. Since level 4 streets present a more homogeneous traffic profiles, it was considered that data from these 5 traffic sensors is representative. The average daily traffic (ADT) volume (presented in Table 9 in the Supplementary material) for each hierarchical street level is similar across months (with the exception of August, December and January). In August, the ADT is lower than for the other months due to the summer holidays, while in December and in January the ADT is higher, possibly due to Christmas related shopping. Based on the traffic volumes collected by the Municipality of Lisbon (Source: CML, Traffic and Urban Mobility Department), when assessing the average traffic volume profile on an hourly basis for each hierarchical street level (see Fig. 4), level 4 streets present a smoother profile, while in level 2 streets the influence of rush hours is clear (Fig. 4a)). Furthermore, when compared to weekdays, in the weekend a more homogenous profile is observed.

2.2. Data analysis In order to assess the energy consumption impacts of shifting traffic from peak hours, the second-by-second driving data was complemented with the traffic volume data, by accounting for average values per driver for each trip when it crossed these sections of road where a traffic sensor is deployed. Using this data, the volume-to-capacity (v/c) ratio was estimated. Volume-to-capacity ratio is calculated as the number of vehicles passing through a traffic sensor in an hour (volume) divided by the historical maximum (capacity) of that traffic sensor (Suresh and Umadevi, 2014). To guarantee the correct estimation of the historical maximum of each traffic counter, outliers were identified and removed from the database. Due to occasional traffic cut off or traffic sensors malfunctioning, very low or null traffic counts occurred 492

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 4. Average traffic volume profile along the day for each hierarchical street level for (a) weekdays and (b) weekends.

on some days and for some traffic sensors. Considering this, daily traffic volumes lower than 1/3 of the average daily traffic volumes for that traffic sensor on that month and considering similar days (weekdays or weekends) were removed from the database. Furthermore, on an hourly basis, values 100% higher than average values for that traffic sensor at that hour were averaged using traffic volumes from the previous and after hours. Consequently, higher values of v/c ratio represent more vehicles passing through each traffic sensor. It should be clear that low values of v/c ratio can occur either if the number of vehicles driving through the street is small or if congestion is happening. For all vehicle driving data passing through a traffic sensor a v/c ratio was attributed. Fig. 5 presents the v/c ratio frequency, showing that the distribution is positively skewed. The left tail (lower percentages of volume-to-capacity ratio) is less representative, with most of the v/c ratio data (91%) occurring between 5% and 55% of the observed maximum traffic volume. In Fig. 5, v/c ratio frequency is presented for the street levels 2, 3 and 4 (no traffic volume data available for level 1). Based on this data, the probability of occurrence of different v/c ratios in each hour of the day was calculated, as presented in Eq. (1). Each hour can present several traffic conditions, with each of these conditions corresponding to different v/c ratios. This probability allows distributing the hourly traffic volume for the tested scenarios, as will be presented in Section 3.2.

Probability of occurrence of different v /c ratios [%]hour i =

Count of occurrences of v /c for hour i Count of ocurrences of v /c for all bins

(1)

In an attempt to identify if a vehicle is driving in free-flow conditions or in congested conditions, the ratio between vehicle speed including idling and vehicle speed excluding idling was calculated for each traffic sensor road segment, per trip and per driver (Eq.

Fig. 5. Volume-to-capacity ratio frequency. 493

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 6. Speed δ frequency.

(2)).

Speed δ [%] =

Speed with idling −1 Speed without idling

(2)

The frequency histogram of this variable is presented in Fig. 6, clearly showing a predominance of two situations: one where these two speeds are equal, which corresponds to a situation where there were no stops; and another where this ratio is < 0%, and consequently traffic congestion may exist. Besides traffic congestion other variables such as crosswalks and traffic signs may also influence this variable, so further analysis will be performed in the future to deepen this analysis. Based on this variable, it was possible to compute the probability of having congested conditions, by using Eq. (3). This probability is based on the frequency of speed δ. When speed δ is lower than 0, then it is probable that any kind of congestion may exist. Consequently, the more occurrences of speed δ lower than 0 the more probable to have congested conditions. This probability was calculated for each time a vehicle crossed a road segment with a traffic sensor deployed.

Probability of having congested conditions [%] =

Count of Speed δ < 0 Count of Speed δ

(3)

In the cases where there was no data available (mainly during the night period), average values were assumed for the probability of having congested conditions. To calculate these average values, similar situations were considered: for rush hours, the average values were calculated from the values of the remaining rush hours; and the same reasoning was applied for non-rush hours. By considering the probability of having congested conditions and its opposite, it was possible to estimate an average circulation speed according to these probabilities. This average speed is a function of the average speed in congested conditions and of the average speed in free-flow conditions, for each v/c ratio, according to the probability of having congested conditions (Eq. (4)). km

Average speed ⎡ h ⎤ ⎣ ⎦ = Probability of having congested conditions × average speed with idling + (1−Probability of having congested conditions ) × average speed without idling

(4)

To quantify the vehicles’ energy consumption in a comparable way, the vehicle Specific Power (VSP) methodology was used (Jiménez-Palacios, 1999). Since different vehicles in the data sample can cross different areas, the influence of the vehicle was excluded by considering an average representative vehicle of the Portuguese fleet (Duarte, 2013). In the VSP methodology, vehicle energy consumption and emissions are obtained from measured data on a second-by-second basis, which are grouped into bins, according to the power requirements. To estimate the power demand by the vehicle, the VSP methodology combines speed, acceleration and road grade, according to Eq. (5) which is applicable to light-duty vehicles (JiménezPalacios, 1999): 494

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Table 3 VSP modes and corresponding power requirements. VSP mode

VSP (W/kg)

VSP mode

VSP (W/kg)

1 2 3 4 5 6 7

VSP < −2 −2 ≤ VSP < 0 0 ≤ VSP < 1 1 ≤ VSP < 4 4 ≤ VSP < 7 7 ≤ VSP < 10 10 ≤ VSP < 13

8 9 10 11 12 13 14

13 ≤ VSP 16 ≤ VSP 19 ≤ VSP 23 ≤ VSP 28 ≤ VSP 33 ≤ VSP VSP ≥ 39

< < < < < <

16 19 23 28 33 39

Table 4 Characteristics of the considered vehicles (based on: Martins (2016)).

ICE displacement (cc) ICE compression ratio ICE Power (kW/rpm) ICE Torque (Nm/rpm) Vehicle mass (kg) Power-to-Weight ratio (W/kg)

VSP [W/kg] =

Power = Mass

d dt

Gasoline

Diesel

998 10.5:1 50.7/6000 93/3600 950 53

1995 16.5:1 85/4000 260/1750 1385 61

(Ekinetic + Epotential ) + Frolling ·v + Faerodynamic ·v m

= v (1.1·a + 9.81·grade + 0.132) + 3.02 × 10−4 ·v 3

(5)

where Ekinetic is kinetic energy, Epotential is potential energy, Frolling is rolling resistance force, Faerodynamic is aerodynamic resistance force, v is speed (m/s), m is mass (kg), a is acceleration (m/s2) and grade is road grade (m/m). Each second of driving is associated to a VSP bin, as presented in Table 3, which shows the 14 VSP modes and the corresponding power requirements intervals for each mode. Using the collected second by second data on speed, acceleration and road grade, VSP was calculated and combined with a fuel consumption curve per VSP mode to quantify energy consumption. In this work, the measured fuel consumption from two representative vehicles (one gasoline and one diesel as presented in Table 4) was considered to guarantee representative results for the average Portuguese fleet for fuel consumption (Duarte, 2013). The selected vehicles were chosen according to the average engine displacement of the Portuguese fleet (76% of the gasoline vehicles have an engine displacement lower than 1.4 l; as for diesel vehicles, 69% of the vehicles have an engine displacement between 1.4 and 2.0 l) (Autoinforma and ACAP, 2000 to 2013). Furthermore, these vehicles follow the Euro 5 standard which corresponds to more than a quarter of the Lisbon fleet, being the most representative Euro standard1 (Autoridade de Supervisão de Seguros e Fundos de Pensões, 2017). These two vehicles were combined in a 57:43 ratio (Autoinforma and ACAP, 2000 to 2013), correspondent to the gasoline to diesel share in the Portuguese fleet. This representative vehicle was used to estimate the average energy consumption, by considering their energy consumption rate per VSP mode, as presented in Table 5. It should be underlined that the methodology presented in this study is prepared to include data from any vehicle analyzed under the VSP methodology. 2.3. Approach for shifting traffic To evaluate the impacts in energy consumption of shifting traffic volumes from rush hours to non-rush hours, an increasing percentage of traffic was moved from rush hours to adjacent non-rush hours. To perform this analysis, reductions in traffic volumes during rush hours (from 8 to 9 h in the morning period and from 17 to 19 h in the afternoon period) were assessed (see Fig. 7). In order to maintain the overall mobility level of service, the vehicles not circulating during rush hours were considered to circulate in the 2 h immediately before and after the rush hours. It was also assumed that traffic volumes would not exceed the observed maximum for that type of road. For this reason reductions in traffic volumes of up to 40% with intervals of 5% were tested. Table 6 presents the variations in traffic volumes for the tested traffic reductions comparing with the baseline case (current situation). For lower traffic reductions (below 30% for the morning period and below 15% in the afternoon), it was possible to shift the traffic to the first hour immediately before and after the rush hours without exceeding the observed maximum and, therefore, the traffic volumes were kept unchanged in the second hour before and after. Furthermore, it should be underlined that equal shares of traffic shift were considered for the before and after hours. For example, if a 200 vehicles reduction would occur during the 8–9 a.m. time period, 100 of these vehicles would circulate at 7 a.m. while the other 100 vehicles would circulate at 10 a.m. The imposed changes in traffic volumes results in expected variations in average speed. Considering this, new average speeds were estimated for both rush and non-rush hours according to the new traffic volumes. Subsequently, the established correlation between energy consumption and speed (see Section 3.3) was used to estimate the average fuel consumption, enabling the assessment 1

Information on the Lisbon fleet share per Euro standard is available; however, no data on average engine displacement is available at a city level.

495

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Table 5 VSP modes and corresponding energy consumption. VSP mode

Energy consumption (MJ/s)

1 2 3 4 5 6 7 8 9 10 11 12 13 14

0.004 0.006 0.007 0.020 0.027 0.037 0.046 0.054 0.064 0.073 0.088 0.097 0.107 0.116

Fig. 7. Representation of the hours of the day for which shifts in traffic volumes were considered (rush hours in red; non-rush hours in green). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) Table 6 Variations in traffic volumes for the tested traffic reductions during rush hours (in grey) comparing with the baseline case.

Traffic reductions

Hour 06:00 07:00 08:00 09:00 10:00 11:00 … 15:00 16:00 17:00 18:00 19:00 20:00 21:00

5%

0%

8%

-5%

-5%

6%

0%

…

0%

8%

-5%

-5%

-5%

10%

0%

10%

0%

16%

-10%

-10%

13%

0%

…

0%

16%

-10%

-10%

-10%

20%

0%

15%

0%

24%

-15%

-15%

19%

0%

…

3%

17%

-15%

-15%

-15%

30%

7%

20%

0%

33%

-20%

-20%

26%

0%

…

8%

17%

-20%

-20%

-20%

40%

15%

25%

0%

41%

-25%

-25%

33%

0%

…

16%

17%

-25%

-25%

-25%

41%

31%

30%

29%

49%

-30%

-30%

19%

10%

…

20%

17%

-30%

-30%

-30%

41%

56%

35%

33%

57%

-35%

-35%

23%

12%

…

20%

17%

-35%

-35%

-35%

41%

89%

40%

38%

65%

-40%

-40%

26%

14%

…

20%

17%

-40%

-40%

-40%

41%

122%

496

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

of the impacts in energy consumption. The average vehicle efficiency was weighted by the number of vehicles to incorporate both the increasing and decreasing energy consumption effects, and therefore to obtain a global indicator of the impacts of shifting the traffic volumes. Data regarding traffic volumes were not available for level 1 streets, therefore only level 2, 3 and 4 streets were considered when performing the connection between vehicle dynamics and traffic conditions and also for the quantification of energy consumption impacts of shifting traffic. Furthermore, due to the lower amount of acquired data for level 4 streets (lower number of traffic sensors are implemented in these streets) these analyses were performed for all levels together. However, when considering the vehicle driving data (see Section 3.1), this analysis is presented separately for each of the four street levels in order to better characterize the impacts of different types of road infrastructure on vehicle dynamics and, consequently, better understanding the potential impacts in energy consumption. It must be noticed that the analyses were performed considering different levels of data aggregation. For instance, to characterize the vehicle dynamics average values per hour were considered while, regarding v/c ratio and speed δ average, hourly values were used but considering only each section of road where a traffic sensor was deployed. 3. Results and discussion According to the objectives of this research work, this section is divided in four areas. Firstly, a characterization of the vehicle dynamics considering speed and acceleration profiles is performed. Secondly, the vehicle dynamics data is correlated with traffic. Afterwards, the impacts on power requirements and energy consumption are analyzed. Finally, the impacts of shifting traffic from rush hours to non-rush hours in energy consumption are assessed. 3.1. Vehicle data analysis The gathered vehicle monitoring data allowed characterizing the average vehicle speed along the day for each hierarchical street level (see Fig. 8). Since this analysis uses only second-by-second driving data, all street levels were considered (from level 1 to level 4). Level 3 and level 4 streets present similar profiles, even presenting very close daily average speeds of 19.4 km/h for level 3 and 18.7 km/h for level 4 streets. As for level 2 streets, the average speed is of 27.5 km/h, which is approximately 50% higher than for level 4. For level 1 streets, the profile is much higher with an average driving speed of 59 km/h. Furthermore, as a result of being streets comparable to highways, level 1 streets not only present higher average speed values but also a higher impact on speed during rush hours. The average speed during rush hours for level 1 streets decreases by almost 20% when comparing with non-rush hours. As for level 2 and 3 streets, these streets present a similar trend, with the average speed decreasing by 16% and 14%, respectively, in rush hour periods. Finally, as expected, due to its characteristics, with lower levels of mobility, in level 4 streets the average speed is less influenced by rush hours, with a decrease of only 5% in the average speed for those periods. Regarding the comparison of morning and afternoon rush hours, during the afternoon rush hours for all street levels a higher decrease in the average speed is observed.

Fig. 8. Average speed profile along the day for each hierarchical street level (Error bars for 95% Confidence Interval). 497

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 9. Average positive (a) and negative (b) acceleration profiles along the day for each hierarchical street level (Error bars for 95% Confidence Interval).

When comparing the average speed profiles between weekdays and weekend days (see Fig. 20 in the Supplementary material) it is visible that for weekdays the influence of rush hours is clear while for weekend days the speed profile does not show that specific trend. Overall, it is possible to conclude that the average speed for weekend days is always higher than for weekdays and that the difference decreases as the street becomes more local. For level 1 streets, the average speed is 20% higher for weekend days; for level 2 18% higher; for level 3 17%; and finally for level 4 it is 10% higher than for weekdays. Fig. 9 shows the vehicles’ average positive and negative acceleration profiles along the day for each hierarchical street level. Even though level 1 streets present the higher average speed, in terms of acceleration, both positive and negative, this type of streets has less aggressive profiles (meaning that harder acceleration/braking events occur less in this type of streets). Level 1 streets are comparable to highways having less variable circulation conditions and lower influence from road infrastructure. These roads are characterized by higher number of lanes, higher lane width and higher posted speed limits. Additionally, level 1 streets have lower number of intersections and no interaction with pedestrians, therefore these characteristics lead to smoother driving profiles with lower average acceleration/braking values (Wang, 2006). As for level 2, 3 and 4 streets, acceleration levels are higher possibly due to the high number of intersections, interaction with pedestrians and influence of traffic signs. Furthermore, although there is a slight influence of rush hours in level 1 acceleration profiles, it is not as clear as for the average speed profile. The combined assessment of speed and acceleration is documented in the literature as an input for vehicle emissions models (Ahn et al., 2002). Traffic simulation models can estimate both vehicle speed and acceleration. However, the usage of such information may lead to significant errors in emissions estimates due to misleading representation of second-by-second driving behaviors (Song et al., 2012). Therefore, real-world data can provide useful information that is crucial to calibrate and bound microscale traffic

Fig. 10. Average acceleration versus average speed for each hierarchical street level. 498

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 11. Average acceleration as a function of slope and speed.

simulation models. Fig. 10 provides information on acceleration versus speed, for each hierarchical street level. It allows verifying that for higher speeds the average acceleration tends to converge. This occurs mainly in level 1 streets. For streets with lower levels of mobility (such as level 4 streets), speed decreases and aggressive events (e.g. extreme accelerations and braking) are more frequent. When simultaneously analyzing acceleration, slope and speed (see Fig. 11), it is possible to observe that negative accelerations mostly occur for negative slopes at low speeds (up to 40 km/h), while positive accelerations occur for higher slopes and at higher speeds (ranging from 35 to 75 km/h). Also, typically, higher speeds are reached in null or lower slopes (both positive and negative). In steeper environments, drivers tend to drive slower either due to vehicle power restrictions (positive slopes) or due to safety precaution (negative slopes). From this vehicle data analysis, it is possible to conclude, that both speed and acceleration profiles are influenced by the type of road, type of day and hour of the day. Typically, these analyses are performed based on infrastructure characterization data without using real word driving data, which provide much more accurate characterization of infrastructure usage as presented in this analysis.

3.2. Connecting vehicle dynamics with traffic conditions Using v/c ratio and speed δ, the influence of traffic in vehicles speed was assessed. Fig. 12 shows that for v/c ratios higher than 25% and with increasing traffic, the average speed decreases. This decrease is higher in less congested conditions (speed δ = 0). The influence of increasing traffic is clearly visible in the decrease of the average speed δ variable, as presented in the blue line in Fig. 12. It is also noticeable from Fig. 12, as expected, that for speed δ lower than zero the average speed is always lower than for speed δ equal to zero. On average, speed with idling (speed δ < 0) is ∼45% lower than speed without idling (speed δ = 0). Based on the traffic data for the city of Lisbon that was coupled with the collected driving data (see Section 2), the average hourly vehicle distribution in absolute values was assessed, as presented in Fig. 13. This distribution shows that the morning rush hours occurs between 8 h and 9 h while in the afternoon, the traffic peak is distributed in a longer period of time, from 17 h to 19 h. Additionally, the probability of occurrence of v/c ratio in each hour of the day was calculated (see Table 7). From Table 7 it is noticeable that, during rush hours, there is a higher probability (represented in shades of red in Table 7) of having intermediate percentages (30–75%) of v/c ratio. In the other hours, there are fewer vehicles circulating and, consequently, there is a higher probability of the road having low v/c ratio (< 30%). High v/c ratio occurrences (> 75%) are rare as they require high traffic volumes without significant congestion in order to have many vehicles passing through the traffic counters. Using the speed δ, it is possible to infer about the probability of having congested conditions. For the same traffic conditions, the higher the ratio between the count of speed δ lower than zero and the count of speed δ lower than or equal to zero, the higher the probability of having congested conditions (see Eq. (3)). Table 8 shows that typically a higher probability of congested conditions occurs during the morning rush hours and in higher traffic volumes. Furthermore, it is also noticeable that for lower percentages of v/ c ratio the probability of having congested conditions can also be high. As previously mentioned, low percentages of v/c ratio can effectively represent a situation of low traffic volume (few vehicles circulating) or it can also be a result of congestion. In congested conditions the traffic does not flow (or flows at very low speeds) and, therefore, the number of vehicles passing by the traffic sensors is lower. 499

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 12. Average speed and speed δ for different percentages of volume-to-capacity ratio (Error bars for 95% Confidence Interval).

Fig. 13. Average hourly traffic distribution in absolute values.

It must be noticed that the probability of having congested conditions is deeply related to the number of occurrences that were monitored for each situation (it is based on the frequency of speed δ). In this sense, especially during the night period (from midnight to 6 a.m.), lower number of occurrences were registered leading to some variability in the probabilities (values of 100% followed by 0%).

3.3. Quantification of energy consumption In order to quantify energy consumption associated to these circulation profiles, the power required to the vehicle on each second was estimated. Fig. 14 presents the average VSP profile, as well as the disaggregation by the 3 hierarchical levels considered. Level 3 and 4 present similar power requirements, on average 40% lower compared to level 2 streets. During rush hours, VSP decreases on average by more than 50% for all levels. 500

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Table 7 Probability of occurrence of volume-to-capacity ratio in each hour (rush hours in grey; in each hour green represents low probability, yellow means intermediate probability and red represents the higher probability).

Table 8 Probability of having congested conditions (rush hours in grey; in each hour green represents low probability, yellow means intermediate probability and red represents the higher probability).

*As referred in Section 2.2, due to lack of data, mainly during the night period, for the cases where data was missing, average values were assumed for the probability of having congested conditions.

The analysis of VSP considering an hourly disaggregation (presented in Fig. 15) clearly shows that the two rush hour periods have higher percentage of time spent at VSP mode 3, which represents idling. During rush hours, traffic volumes are higher and consequently the probability of congestion occurring is also higher. In this sense, the higher percentage of time spent idling during rush hours may be either due to stops in traffic signs or due to congested traffic. Nevertheless, when analyzing this data separately for each street level (see Fig. 16), differences among them are noticeable. For level 2 both morning and afternoon rush hours have a high percentage of time spent on VSP mode 3. However, the afternoon rush hour importance decreases as the streets become more local. For level 4 streets the increase in the percentage of time spent on VSP mode 3 in the afternoon rush hours is almost unnoticeable. This denotes the impact of traffic congestion that occurs mainly on streets with higher levels of mobility (such as level 2 streets). The impacts on local streets during afternoon peak hours are less visible possibly due to the higher distribution of traffic along time. While in the morning, the traffic peak is more concentrated around 8 h, in the afternoon it is concentrated from 17 h to 19 h. The analysis of the VSP distribution as a function of traffic volume is also essential to understand how energy consumption may 501

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 14. Average VSP profile along the day.

Fig. 15. Percentage of time spent in each VSP mode as a function of hour of the day for all hierarchical street levels.

evolve (see Fig. 17a). The increase in percentage of time spent in VSP mode 3 is clearly visible as traffic volume increases. For traffic volumes above 1100 vehicles per hour, the percentage of time spent in VSP mode 3 surpasses 40%. However, it is also clear that even with low traffic volumes the percentage of time spent idling is never lower than 30% which possibly represents the effect of traffic signs. These results indicate that traffic signs have a major influence on traffic flows and deeply impact the driving dynamics even in free-flow conditions. The same conclusions can be withdrawn from Fig. 17b, where for lower speeds there is a higher percentage of time spent in VSP mode 3. Assuming a representative vehicle (resulting from a combination of current gasoline and diesel vehicles corresponding to the average national fleet fuel distribution) it is possible to estimate energy consumption in a comparable way. This means that the vehicle dynamic sections are assumed to be driven with the same average vehicle, to eliminate the influence of this variable. A correlation between driving speed and energy consumption was obtained by using the VSP methodology (as presented in Fig. 18). 502

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 16. Percentage of time spent in each VSP mode as a function of hour of the day for: (a) Level 2; (b) Level 3; (c) Level 4.

Using the VSP methodology enabled associating average fuel consumption to each VSP mode which was analyzed for several speed bins in intervals of 5 km/h. For the purpose of this analysis, aggregated data regarding all street levels were considered. The established correlation is shown in Fig. 18 and is in accordance with previous results in the literature (André and Hammarström, 2000; Gkatzoflias et al., 2007; Samaras and Ntziachristos, 1998). However, the presented correlation is based on real-world driving data and demonstrates the usefulness of the application of the VSP methodology. This method allows performing large-scale driving monitoring campaigns without the need of measuring energy consumption and pollutants emissions. Using Eq. (6) (established in Fig. 18), it is possible to estimate energy consumption based on speed variations (in km/h):

l Average consumption ⎡ km⎤ = 0.003 × (speed)2−0.358 × speed + 15.741 ⎣ 100 ⎦

(6)

3.4. Impacts of shifting traffic in energy consumption The main objective of this work was to estimate the impacts in energy consumption of reducing traffic volume in rush hours but by maintaining the same mobility level of service. As a result of the established traffic variations (see Table 6 in Section 2.3), new vehicle distributions were estimated which enabled the estimation of new average circulation speeds (see Fig. 21 in the Supplementary material, as an example for a 20% traffic shift). Fig. 19 shows the changes in average speed for each percentage of traffic shift. Overall, for percentages of traffic shift up to 35% there is an increase in the average speed comparing to the current situation. However, while for traffic shift of up to 20%, the average speed variation has an increasing trend, for higher than 20% traffic shift, that trend overturns and the difference on average speed starts to decrease. For percentages of traffic shift equal to 35% or higher, there are no benefits in shifting traffic. Probably, this is due to the fact that the non-rush hours adjacent to the rush hours become overloaded with the vehicles that shifted from the rush hours. The higher increase in the average speed (∼0.6%) can be obtained for a traffic shift of 20%, while the lowest positive variation in the average speed (∼0.2%) occurs for a 30% traffic shift percentage. Subsequently, by applying Eq. (6) to the obtained speeds, it was possible to estimate the new energy consumption. Fig. 19 also 503

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 17. Percentage of time spent in each VSP mode as a function of traffic volume (a) and of average speed (b) for all considered street levels.

shows the obtained reductions in the average energy consumption for each percentage of traffic shift. The average energy consumption variation follows the same trend as the average speed but in the opposite direction, i.e. instead of increasing, the average energy consumption decreases. Overall, the maximum reduction in energy consumption (∼0.4%) is achieved for a 20% traffic shift while the lowest reduction (∼0.1%) occurs for a 30% traffic shift percentage. As observed for the average speed, for percentages of traffic shift above 20% the positive impact in terms of energy consumption reduction starts to decrease. For a traffic shift of 40% the overall performance in terms of energy consumption worsens comparing to the current situation. As previously referred, possibly this occurs due to the non-rush hours becoming overloaded with the vehicles that shifted from the rush hours. Globally, by shifting 20% of the rush hours traffic to the adjacent non-rush hours – the best case scenario – trip time can be reduced by 1.4% (considering only the average distance – 0.51 km – travelled on the section of road with a traffic sensor). Additionally, savings of up to 6 l of fuel and avoided CO2 emissions of 14.5 kg may be achieved per day. 4. Conclusions The importance of the road transport sector to energy consumption and emissions justifies the need for a more accurate analysis on how to overcome its impacts. Consequently, the possibility to monitor vehicle performance in real world conditions opens new 504

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Fig. 18. Average consumption as a function of average speed.

Fig. 19. Variations (%) in average speed (km/h) and in energy consumption (based on the l/100 km values).

opportunities for assessment of alternative options. This work is an example of the potentialities of using this type of data, by connecting traffic and vehicle dynamic data in an innovative approach, in order to assess the impacts of shifting traffic from peak hours, applied to a sample of 40 drivers monitored in the city of Lisbon for a period of six months. The developed methodology is replicable elsewhere, given that the real world data is available. Furthermore, this study provides useful insights on the mobility characteristics that will make possible to develop new systems (e.g. Pay-As-You-Drive and Peak-hour tolls) and policy measures, bringing unquestionable benefits in terms of decreases in fuel consumption and exhaust emissions. Traffic congestion is a mobility challenge for which no easily applicable measures are available, either politically and physically and financially. Therefore, deepening knowledge on this issue must be pursued in future works. In terms of the potential gains associated to shifting traffic from peak hours, reductions on energy consumption from 0.1% to 0.4% can be obtained for traffic volumes shifts from 5 to 30%. Overall, the maximum reduction in energy consumption (∼0.4%) is achieved for a 20% traffic shift. Furthermore, the results allow concluding that for traffic shifts above 35% there is a worsening of 505

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

energy consumption comparing to the current situation. Probably, this is due to the fact that the non-rush hours adjacent to the rush hours become overloaded with the vehicles that shifted from the rush hours. Regarding the average speed variation, it follows the same trend as the average energy consumption but in the opposite direction, i.e. instead of decreasing, the average speed increases. Globally, for the best case scenario, considering only the sections of roads with traffic sensors, a 1.4% reduction in trip time may be achieved, as well as savings of up to 6 l of fuel and 14.5 kg of avoided CO2 emissions per day. In conclusion, these results show that positive impacts may be achieved through the implementation of measures that promote the shift of traffic volumes from rush hours to the adjacent non-rush hours. However, when considering the promotion of public policies to decrease traffic congestion during rush hours a thoughtful assessment of the impacts in energy consumption must be performed in order to avoid the implementation of measures leading to an increase in energy consumption. An integrated strategy that enables proper enforcement or monitoring of the actual energy impacts should also be promoted. Acknowledgments The authors acknowledge Fundação para a Ciência e Tecnologia through support from: Doctoral grant (PD/BD/105714/2014); IDMEC, under LAETA (Project UID/EMS/50022/2013); IN+ Strategic Project (Project UID/EEA/50009/2013); and the Suscity Project (MITP-TB/C S/0026/2013). The authors would also like to acknowledge the i2d Project for providing the on-board data loggers and Direcção Municipal de Mobilidade e Transportes (DMMT/DGMT/DGT) of Câmara Municipal de Lisboa for providing the traffic count data. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.trd.2018.03. 008. References Ahn, K., et al., 2002. Estimating vehicle fuel consumption and emissions based on instantaneous speed and acceleration levels. J. Transp. Eng. 128 (2), 182–190. http://dx.doi.org/10.1061/(ASCE)0733-947X(2002) 128:2(182). André, M., Hammarström, U., 2000. Driving speeds in Europe for pollutant emissions estimation. Transp. Res. Part D: Transp. Environ. 5 (5), 321–335. http://dx.doi. org/10.1016/S1361-9209(00)00002-X. André, M., et al., 1995. Driving cycles for emission measurements under European conditions. SAE Technical Paper 950926. https://doi.org/10.4271/950926. Autoinforma, ACAP, 2000 to 2013. Estatísticas do sector automóvel. Autoridade de Supervisão de Seguros e Fundos de Pensões, 2017. Parque Automóvel Seguro - 2016, Ano de construção (in Portuguese). http://www.asf.com.pt/NR/ exeres/7D383D46-9431-416E-98C7-395B0A9E7080.htm (20 October 2017). Brundell-Freij, K., Ericsson, E., 2005. Influence of street characteristics, driver category and car performance on urban driving patterns. Transp. Res. Part D: Transp. Environ. 10 (3), 213–229. http://dx.doi.org/10.1016/j.trd.2005.01.001. Câmara Municipal de Lisboa, 2005. Lisboa: O Desafio da Mobilidade (in portuguese), Colecção de Estudos Urbanos – Lisboa XXI – 7, p. 291. de Vlieger, I., et al., 2000. Environmental effects of driving behaviour and congestion related to passenger cars. Atmos. Environ. 34 (27), 4649–4655. http://dx.doi. org/10.1016/S1352-2310(00)00217-X. Della Ragione, L., Meccariello, G., 2017. Statistical approach to identify naples city’s real driving cycle referring to the worldwide harmonized light duty test cycle (WLTC) framework. WIT Trans. Ecol. Environ. 210, 555–566. http://dx.doi.org/10.2495/SDP160461. Duarte, G., 2013. A Methodology to Estimate Vehicle Fuel Consumption and Pollutant Emissions in Real-World Driving based on Certification Data. Mechanical Engineering. Technical University of Lisbon. Ericsson, E., 2001. Independent driving pattern factors and their influence on fuel-use and exhaust emission factors. Transp. Res. Part D: Transp. Environ. 6 (5), 325–345. http://dx.doi.org/10.1016/S1361-9209(01)00003-7. Commission, European, 2015. EU Transport in Figures: Statistical Pocketbook 2015. Publications Office of the European Union, Luxembourg. Federal Highway Administration, 2013. Highway Functional Classification: Concepts, Criteria and Procedure. U.S. Department of Transportation. Fontaras, G., et al., 2017. Fuel consumption and CO2 emissions from passenger cars in Europe-Laboratory versus real-world emissions. Prog. Energy Combust. Sci. 60, 97–131. http://dx.doi.org/10.1016/j.pecs.2016.12.004. Geenhuizen, M., 2009. ICT applications on the road to sustainable urban transport. Eur. Transp./Trasporti Europei 41, 47–61. Gkatzoflias, D., et al., 2007. COPERT 4: Computer Programme to Calculate Emissions from Road Transport – User Manual (Version 5.0). European Environment Agency. Gonder, J., et al., 2007. Using global positioning system travel data to assess real-world energy use of plug-in hybrid electric vehicles. Transp. Res. Rec.: J. Transp. Res. Board 2017, 26–32. http://dx.doi.org/10.3141/2017-04. Google, 2016. Geocoding API - Getting started. https://developers.google.com/maps/documentation/geocoding/start (15 September 2016). Hallmark, S.L., et al., 2002. Characterizing on-road variables that affect passenger vehicle modal operation. Transp. Res. Part D: Transp. Environ. 7 (2), 81–98. http:// dx.doi.org/10.1016/S1361-9209(01)00012-8. iTds and IDMEC-IST, 2011. i2d - intelligence to drive, Driving efficiency (I&D project). Jensen, S.S., 1995. Driving patterns and emissions from different types of roads. Sci. Tot. Environ. 169 (1–3), 123–128. http://dx.doi.org/10.1016/0048-9697(95) 04640-M. Jiménez-Palacios, J.L., 1999. Understanding and Quantifying Motor Vehicle Emissions with Vehicle Specific Power and TILDAS Remote Sensing. Department of Mechanical Engineering, Massachusetts Institute of Technology. Kompfner, P., et al., 2008. ICT for Clean & Efficient Mobility Final Report, Safety Forum. LiveDrive, 2015. i2D Telematics - How does it work. https://www.i2d.co/i2dpubportal/portal/tec.xvw (January 2016). Martins, H.A.C., 2016. Definition and Evaluation of Representative Driving Cycles of Real World Conditions in Different Urban Contexts. University of Lisbon, IST. Pelkmans, L., Debal, P., 2006. Comparison of on-road emissions with emissions measured on chassis dynamometer test cycles. Transp. Res. Part D: Transp. Environ. 11, 233–241. http://dx.doi.org/10.1016/j.trd.2006.04.001. Samaras, Z., Ntziachristos, L., 1998. Average hot emission factors for passenger cars and light duty trucks In: The Project: Methodologies for estimating air pollutant emissions from transport (MEET) – Report 7. Lab. of Applied Thermodynamics, Aristotle Univ., Thessaloniki, Greece. Sevtsuk, A., Ratti, C., 2010. Does urban mobility have a daily routine? Learning from the aggregate data of mobile networks. J. Urban Technol. 17 (1), 41–60. http:// dx.doi.org/10.1080/10630731003597322.

506

Transportation Research Part D 62 (2018) 489–507

M. Faria et al.

Smidfelt-Rosqvist, L., 2003. On the Relations Between Driving Patterns, Exhaust Emissions and Network Characteristics in Urban Driving. Lund University. Song, G., et al., 2012. Applicability of traffic microsimulation models in vehicle emissions estimates: Case study of VISSIM. Transp. Res. Rec.: J. Transp. Res. Board (2270), 132–141. http://dx.doi.org/10.3141/2270-16. Suresh, V., Umadevi, G., 2014. Empirical methods of capacity estimation of urban roads. Global J. Res. Eng. 14 (3). Tzirakis, E., et al., 2006. Vehicle emissions and driving cycles: comparison of the Athens driving cycle (ADC) with ECE-15 and European driving cycle (EDC). Glob. NEST J. 8 (3), 282–290. Tzirakis, E., et al., 2007. Impact of driving styles on fuel consumption and exhaust emissions: Defensive and aggressive driving style. In: 10th International Conference on Environmental Science and Technology. Wang, J., 2006. Operating Speed Models for Low Speed Urban Enviroments based on In-Vehcile GPS. Georgia Institute of Technology. Wang, P., et al., 2012. Understanding road usage patterns in urban areas. Sci. Rep. 2, 1001. Wang, Q., et al., 2008. Characterization of vehicle driving patterns and development of driving cycles in Chinese cities. Transp. Res. Part D: Transp. Environ. 13 (5), 289–297. http://dx.doi.org/10.1016/j.trd.2008.03.003. Zhang, S., et al., 2014. Real-world fuel consumption and CO2 (carbon dioxide) emissions by driving conditions for light-duty passenger vehicles in China. Energy 69, 247–257. http://dx.doi.org/10.1016/j.energy.2014.02.103.

507