University of California Transportation Center
UCTC- FR- 2010- 29
A Study on Potential Environmental Benefits of Green Driving
Strategies with NGSIM Data
Wen- Long Jin, Daji Yuan,
and Hao Yang
University of California, Irvine
August 2010
1Author for correspondence
A STUDY ON POTENTIAL ENVIRONMENTAL BE 1 NEFITS OF GREEN DRIVING
2 STRATEGIES WITH NGSIM DATA
3
4
5 WEN- LONG JIN1
6 Assistant Professor
7 Department of Civil and Environmental Engineering
8 Institute of Transportation Studies
9 University of California, Irvine
10 Irvine, CA 92697- 3600
11 Email: wjin@ uci. edu
12
13 DAJI YUAN
14 Ph. D. Student
15 Department of Civil and Environmental Engineering
16 Institute of Transportation Studies
17 University of California, Irvine
18 Irvine, CA 92697- 3600
19 Email: dajiy@ uci. edu
20
21 HAO YANG
22 Ph. D. Student
23 Department of Civil and Environmental Engineering
24 Institute of Transportation Studies
25 University of California, Irvine
26 Irvine, CA 92697- 3600
27 Email: hyang5@ uci. edu
28
29
30
31 Word Count: 4000+ 250×12= 7000
32 August 1, 2010
33 SUBMITTED TO 2011 TRB ANNUAL MEETING
2
34 Abstract
35 The main purpose of this paper is to examine potential environmental benefits of green driving
36 strategies with NGSim data on Interstate- 80 near Berkeley, California. We calculate vehicles
37 emissions before and after applying green driving strategies with the VT- Micro emission model.
38 For each vehicle, its trajectory before applying green driving strategies is observed and given in
39 the dataset. We assume that, with the help of green driving strategies, the vehicle could drive at a
40 constant speed over the whole road section with the same travel distance and time as before.
41 After examining impacts of speed- acceleration adjustment on calculated emissions and fuel
42 consumptions, we choose 5127 out of 9951 cars and estimate potential savings in HC, CO, NOx,
43 CO2, and fuel consumptions. With a new model of the relationships between emissions/ fuel
44 consumptions and average speeds, we can fit the data with R- squares close to or greater than 0.9
45 and find that green driving strategies are most effective for traffic flows with average speeds
46 around 50 km/ h and potential savings can be from 20% to 60% for different pollutants. In the
47 future, we will continue our studies with more realistic information on vehicle types and other
48 emission models.
49
50 Keywords: Green Driving, NGSIM Data, VT- Micro, Vehicle Emissions, Traffic Conditions
51
3
52 1 INTRODUCTION
53 Nearly one third of the greenhouse gas ( GHG) emissions are from the transportation sector ( 1).
54 Transportation is also a major source of pollutants of Hydrocarbons ( HC), Carbon Monoxide
55 ( CO), Nitrogen Oxides ( NOX). Although alternative fuel or electric vehicles have potential to
56 reduce vehicle emissions significantly, their benefits cannot be felt in the near future. In contrast,
57 we could apply technologies such as adaptive cruise control ( ACC) and inter- vehicle
58 communications ( IVC) to manage traffic conditions so as to avoid stop- and- go traffic and
59 therefore to reduce emissions and fuel consumptions. That is, we can devise " green driving"
60 strategies to smooth traffic flow.
61 Previous studies have explored the potential of using technologies to smooth traffic in order to
62 reduce vehicle emissions. Bose et al. ( 2) studied the impacts of ACC on vehicle emissions using
63 CMEM. Ahn and Rakha ( 3) attempted to quantify the energy and environmental impacts of route
64 choice decisions using in- field collected global positioning system ( GPS) data with microscopic
65 and macroscopic emission simulations. Their results showed that the shortest path could increase
66 emissions although it could save driver’s travel time. Barth and Boriboonsomsin ( 4) introduced
67 an environmental friendly dynamic eco- driving advice system which relies on real time traffic
68 information provided by the California Freeway Performance Measurement System ( PeMS).
69 This concept was further investigated under real- world driving conditions. Their results showed
70 that the reduction in fuel consumption and emissions for eco- driving vehicles is in the order of
71 10% to 20% compared to non eco- driving vehicles in this real- world experiment. El- Shawarby,
72 Ahn and Rakha ( 5) used on board devices to measure real time energy- consumption and
73 emissions on different levels of cruise speed and acceleration. They concluded that the optimum
74 driving speed range is 60- 90 km/ h for energy- consumption and emissions. Barth and
75 Boriboonsomsin ( 6) used collected GPS trajectory data to investigate real world carbon dioxide
76 ( CO2) emissions in southern California. Through using a microscopic emission estimation model
77 CMEM, they proposed and compared strategies that can reduce CO2 based on different traffic
78 conditions. They did not discuss other emission components such as HC, CO, NOx. Treiber,
79 Kesting and Thiemann ( 7) used NGSIM trajectory data to estimate fuel consumption with a
80 physics fuel consumption model. They smoothed vehicle trajectory data before using them to
81 estimate fuel consumptions. However, they did not use any available microscopic emission
82 models.
83 In this paper, we propose to study potential environmental benefits of green driving strategies
84 with NGSIM trajectory data and the VT- Micro microscopic emission model. In our study, we do
85 not consider implementation issue of green driving strategies. Instead, we assume that all
86 vehicles can drive at a constant speed after green driving. Compare emissions of this ideal
87 scenario with realistic data then can give us insights on when green driving strategies are the
88 most effective. In this study, we will carefully examine the available data sets and calculations of
89 emissions and fuel consumptions. We decide to only choose a portion of vehicles, whose second-90
by- second speeds and acceleration rates are within the feasible domain of the VT- Micro model.
91 In addition to examining potential environmental benefits for individual vehicles, we propose a
92 new model to fit the relationships between emissions/ fuel consumptions and average speeds.
93 With the new model we can then determine the magnitude and conditions of most savings.
94 This paper is organized as follows. In section 2, we describe the VT- Micro model, NGSIM data,
95 and the procedure of computing emissions. In section 3, we examine impacts of trajectory
96 sampling and speed- acceleration adjustment on emissions. In section 4, we discuss potential
4
environmental benefits for individual vehicles and study models for 97 fitting the relationships
98 between emissions/ fuel consumptions and average speeds. In section 5, we summarize this paper
99 and discuss possible directions for follow- up studies.
100
101 2 METHODOLOGY
102 2.1 The VT- Micro Model
103 Microscopic emission models estimate instantaneous emissions and fuel consumption rates for
104 different classes of vehicles based on instantaneous speeds and acceleration rates in vehicle
105 trajectory data. The VT- Micro model was developed from experimentation with numerous
106 polynomial combinations of speed and acceleration levels ( 8). This model was tested using
107 chassis dynamometer data collected from the Oak Ridge National Laboratory ( ORNL). With a
108 relatively good fit to the original data, the final regression model has a combination of linear,
109 quadratic and cubic speed and accelerations. The ORNL data consisted of nine normal emitting
110 vehicles which included six light- duty passenger cars and three light duty trucks.
111 From ( 9), the VT- Micro model can be written as 𝑀𝑀 𝑀𝑀 𝑒𝑒 ( 𝑣𝑣 , 𝑎𝑎 )= 􁉊 𝑒𝑒 Σ Σ ( 𝐿𝐿 𝑖𝑖 , 𝑗𝑗 𝑒𝑒 3 𝑗𝑗
= 0 3 𝑖𝑖
= 0 × 𝑣𝑣 𝑖𝑖 × 𝑎𝑎 𝑗𝑗 ) 𝑓𝑓 𝑓𝑓 𝑎𝑎 ≥ 0 𝑒𝑒 Σ Σ ( 𝑀𝑀 𝑖𝑖 , 𝑗𝑗 𝑒𝑒 3 𝑗𝑗
= 0 3 𝑖𝑖
112 = 0 × 𝑣𝑣 𝑖𝑖 × 𝑎𝑎 𝑗𝑗 ) 𝑓𝑓 𝑓𝑓 𝑎𝑎 < 0  ( 1)
113 where 𝑣𝑣 and 𝑎𝑎 are the instantaneous speed and acceleration rate, respectively, and 𝑀𝑀 𝑀𝑀 is the
114 second- by- second emission rate or fuel consumption. The unit of acceleration is km/ h/ s, and the
115 unit of speed is km/ h. Here 𝐿𝐿 𝑒𝑒 𝑖𝑖 , 𝑗𝑗 and 𝑀𝑀 𝑖𝑖 𝑒𝑒 , 𝑗𝑗 are coefficients for acceleration and deceleration
116 regimes and differ for different pollutants or fuel consumption. In our study, we consider
117 emissions in HC ( 𝑒𝑒 = 1), CO ( 𝑒𝑒 = 2), NOX ( 𝑒𝑒 = 3), CO2 ( 𝑒𝑒 = 4), and fuel consumption
118 ( 𝑒𝑒 = 5). Their units are milligrams/ second ( mg/ s) for vehicle emissions and liter/ second for fuel
119 consumption.
120 The VT- Micro model is only feasible for a speed- acceleration domain described by the following
121 equations: 𝑣𝑣 ≤ 120 122 𝑎𝑎 ≤− 0.00002∗ 𝑣𝑣 3+ 0.0031∗ 𝑣𝑣 2 − 0.2558∗ 𝑣𝑣 + 14.4, ( 2) 123 − 5≤ 𝑎𝑎 ≤ 10.
124 Such a domain is shown with the solid lines in Figure 1. This domain applies to both LDV and
125 LDT.
126 2.2 NGSIM Data
127 We use data sets collected on eastbound I- 80 in the San Francisco Bay area in Emeryville, CA.
128 There are three data sets on April 13, 2005 during three 15- minute periods: 4: 00 pm to 4: 15 pm,
129 5: 00 pm to 5: 15 pm, and 5: 15 pm to 5: 30 pm. The other data set was collected on December 3,
130 2003 during 2: 35 pm. and 3: 05 pm. In this study, we only consider passenger cars, which belong
131 to class 2 in NGSIM data sets, since there are only very few heavy trucks, which belong to class
132 3, and motor- cycles, which belong to class 1. There are 4543 ‘ auto’ vehicle trajectories in the
5
data set of 2: 35pm to 3: 05pm. There are 1942, 1742, and 1724 ‘ auto’ vehicle 133 trajectories in the
134 data set of 4: 00 pm to 4: 15 pm, 5: 00 to 5: 15 and 5: 15 to 5: 30.
135 In the NGSIM data sets, vehicles’ trajectories are provided every one tenth ( 4: 00- 4: 15, 5: 00- 5: 30)
136 or one fifteenth second ( 2: 35- 3: 05), but VT- Micro estimate emissions with second- by- second
137 speeds and acceleration rates. To make the emission estimation consistent, we do not use speeds
138 and accelerations provided by NGSIM. Instead, we used location information provided by
139 NGSIM to calculate second- by- second speeds and accelerations. We first sample second- by-140
second location information from NGSIM. For a vehicle i, if its trajectory spans from 𝑡𝑡 𝑖𝑖 ( 𝑎𝑎 )
141 ( seconds) to 𝑡𝑡 𝑖𝑖 ( 𝑏𝑏 ) ( seconds), we choose the sampling time points as: 𝑡𝑡 𝑖𝑖 ( 0), 𝑡𝑡 𝑖𝑖 ( 0)+ 1,⋯, 𝑡𝑡 𝑖𝑖 ( 0)+ 142 𝑇𝑇 𝑖𝑖 , where the first sampling point 𝑡𝑡 𝑖𝑖 ( 0)∈[ 𝑡𝑡 𝑖𝑖 ( 𝑎𝑎 ), 𝑡𝑡 𝑖𝑖 ( 𝑎𝑎 )+ 1], and 𝑇𝑇 𝑖𝑖 is the total travel time of the
143 vehicle. Then we use this location information to calculate the average speeds and acceleration
144 rates during [ 𝑡𝑡 𝑖𝑖 ( 0)+ 𝑡𝑡 − 1, 𝑡𝑡 𝑖𝑖 ( 0)+ 𝑡𝑡 ] for 𝑡𝑡 = 1,⋯, 𝑇𝑇 𝑖𝑖 as follows 145 𝑣𝑣 𝑖𝑖 ( 𝑡𝑡 )= 3600( 𝑥𝑥 𝑖𝑖 ( 𝑡𝑡 𝑖𝑖 ( 0)+ 𝑡𝑡 )− 𝑥𝑥 𝑖𝑖 ( 𝑡𝑡 𝑖𝑖 ( 0)+ 𝑡𝑡 − 1)) ( 3) 146 𝑎𝑎 𝑖𝑖 ( 𝑡𝑡 )= 𝑣𝑣 𝑖𝑖 ( 𝑡𝑡 + 1)− 2 𝑣𝑣 𝑖𝑖 ( 𝑡𝑡 − 1) ( 4)
147 where we assume that 𝑣𝑣 𝑖𝑖 ( 0)= 𝑣𝑣 𝑖𝑖 ( 1) and 𝑣𝑣 𝑖𝑖 ( 𝑇𝑇 𝑖𝑖 + 1)= 𝑣𝑣 𝑖𝑖 ( 𝑇𝑇 𝑖𝑖 ). Note that here the units of
148 location, speed, and acceleration are km, km/ h, and km/ h/ s, respectively, which are consistent
149 with VT- Micro’s units. Thus the travel distance of the vehicle is given by 𝑑𝑑 𝑖𝑖 = 𝑥𝑥 𝑖𝑖 ( 𝑡𝑡 𝑖𝑖 ( 0)+ 𝑇𝑇 𝑖𝑖 )− 𝑥𝑥 𝑖𝑖 􀵫 𝑡𝑡 𝑖𝑖 ( 0) 􀵯 ,
150 and the average speed is 𝑣𝑣 𝑖𝑖 = 3600 𝑑𝑑 𝑇𝑇 𝑖𝑖 𝑖𝑖 151 . ( 5)
152
153 Figure 1 Relationship between speeds and acceleration rates: Red dots for cars in data sets
154 of 4: 00- 4: 15pm, 5: 00- 5: 15pm, and 5: 15- 5: 30pm; Green plus signs for cars in the data set of
155 2: 35- 3: 05pm; and Solid lines are the feasible speed- acceleration domain for VT- Micro's
156 LDV3 vehicles.
6
Figure 1 shows the distribution of second- by- second speeds and acceleration 157 rates of all vehicles
158 in the four data sets. From the figure, we can see that most speeds and acceleration rates are
159 within the feasible domain of VT- Micro as shown in equation ( 1) for the later three data sets.
160 But for the earliear data set, most of speeds and accelerations are outside of the feasible domain.
161
162 Figure 2. Average speeds of cars in different data sets: The four regions from left to right
163 correspond to data sets of 2: 35- 3: 05, 4: 00- 4: 15, 5: 00- 5: 15, and 5: 15- 5: 30, respectively
164 Figure 2 shows the average speeds of cars in four data sets. From the figure we can clearly see
165 that congestion level increases from the first to the fourth data sets, as expected. From 2: 35 to
166 3: 05, the first 2000 vehicles travel at free flow speed, and it is free flow during this time. But for
167 the rest 2500 vehicles, the average speeds keep decreasing, and it suggests that congestion
168 propagated on the road section. In the latter three data sets, there are two groups of average
169 speeds, where the higher average speeds correspond to cars using the carpool lane. Therefore,
170 these data sets can represent three important regimes of traffic: free flow, formation of
171 congestion, and congested traffic. In real world, there is a fourth regime when congestion
172 dissipates.
173 2.3 Emission Estimation Before and After Green Driving Strategies
174 For vehicle i, we assume that some green driving strategies can completely smooth out its
175 trajectory on the road section. That is, after using green driving strategies, it travels at a constant
176 speed 𝑣𝑣 𝑖𝑖 with a zero acceleration speed. With this assumption, the vehicle's travel distance and
177 time do not change. Therefore here we ignore the impacts of green driving strategies on traffic
178 flow itself. This assumption is quite strong, since it does not consider the interactions among
179 vehicles. However, by comparing vehicle emissions before and after green driving strategies, we
180 will be able to obtain some insights on their potential environmental benefits.
7
Before applying Equation 1 to calculate emissions for a vehicle 181 before and after green driving
182 strategies, we want to adjust its speeds and accelerations so that they fall in the feasible domain
183 of the VT- Micro model. From Figure 1 we can see that only minimal adjustments are applied for
184 congested traffic, but substantial adjustments are necessary for free flow. Since such adjustments
185 would change our results, we need to interpret our results cautiously, in particular for free flow.
186 For vehicle i's speed and acceleration rate at time t, we make the following adjustments to its
187 speed and acceleration before or after green driving, 188 𝑣𝑣 􀷜 𝑖𝑖 ( 𝑡𝑡 )= min ⁡ ( 𝑣𝑣 𝑖𝑖 ( 𝑡𝑡 ), 120), 189 𝑎𝑎 􀷜 𝑖𝑖 ( 𝑡𝑡 )= min ⁡ ( ai( 𝑡𝑡 ), max(− 0.00002 𝑣𝑣 􀷜 𝑖𝑖 ( 𝑡𝑡 ) 3+ 0.0031 𝑣𝑣 􀷜 𝑖𝑖 ( 𝑡𝑡 ) 2 − 0.2558 𝑣𝑣 􀷜 𝑖𝑖 ( 𝑡𝑡 )+ 14.4,0)), ( 6) 190 𝑎𝑎 􀷜 𝑖𝑖 ( 𝑡𝑡 )= max ⁡ ( min( 𝑎𝑎 􀷜 𝑖𝑖 ( 𝑡𝑡 ), 10),− 5) .
191 Obviously for trajectories after green driving, only the first equation above is applied, since 192 𝑎𝑎 𝑖𝑖 ( 𝑡𝑡 )= 0. Note that the second equation above is slightly different from Equation 2, since 193 − 0.00002 𝑣𝑣 𝑖𝑖 ( 𝑡𝑡 ) 3+ 0.0031 𝑣𝑣 𝑖𝑖 ( 𝑡𝑡 ) 2 − 0.2558 𝑣𝑣 𝑖𝑖 ( 𝑡𝑡 )+ 14.4< 0 when speed is greater than 100
194 km/ h. In this case, we set the acceleration as 0. Again, such an adjustment does not reflect the
195 realistic situation. The effects of such adjustment will be discussed in Section 3.
196 In our study we calculate the emission or fuel consumption rate for vehicle 𝑖𝑖 as follows ( 𝑒𝑒 = 197 1,⋯, 5) 198 𝐸𝐸 𝑒𝑒 , 𝑖𝑖 ( 𝒗𝒗 􀷝 𝑖𝑖 , 𝒂𝒂 􀷝 𝑖𝑖 )= Σ 𝑡𝑡 𝑡𝑡 𝑖𝑖 =( 𝑡𝑡 0 𝑖𝑖 )(+ 0 𝑇𝑇 )+ 𝑖𝑖 1 𝑀𝑀 𝑀𝑀 𝑑𝑑 𝐸𝐸 𝑖𝑖 𝑒𝑒 􀵫 𝑣𝑣 􀷜 𝑖𝑖 ( 𝑡𝑡 ), 𝑎𝑎 􀷜 𝑖𝑖 ( 𝑡𝑡 ) 􀵯 , ( 7)
199 which has the unit of mg/ km and l/ km for fuel consumption. We can see that, after green driving,
200 the emission rate is 𝐸𝐸 􀴤 𝑒𝑒 , 𝑖𝑖 = Σ 𝑡𝑡 𝑡𝑡 𝑖𝑖 =( 𝑡𝑡 0 𝑖𝑖 )(+ 0 𝑇𝑇 )+ 𝑖𝑖 1 𝑑𝑑 𝑀𝑀 𝑖𝑖 𝑀𝑀 𝐸𝐸 𝑒𝑒 ( 𝑣𝑣 𝑖𝑖 ,0)= 𝑀𝑀 𝑀𝑀 𝐸𝐸 𝑒𝑒 𝑑𝑑 ( 𝑣𝑣 𝑖𝑖 𝑖𝑖 ,0) 𝑇𝑇 𝑖𝑖 = 3600 𝑀𝑀 𝑀𝑀 𝐸𝐸 𝑣𝑣 𝑒𝑒 ( 𝑖𝑖 𝑣𝑣 􀷜 𝑖𝑖 ,0) 201 ( 8)
202 Then the potential saving percentage of green driving strategy for vehicle i and pollutant e can be
203 calculated by 𝑆𝑆 𝑒𝑒 , 𝑖𝑖 = 1− 𝐸𝐸 𝐸𝐸 􀴤 𝑒𝑒 𝑒𝑒 ,, 𝑖𝑖 𝑖𝑖 204 . ( 9)
205 3 PRELIMINARY RESULTS
206 In this section, we examine the impacts of trajectory sampling and speed- acceleration adjustment
207 on emission calcualtions.
208 3.1 Sensitivity of Emissions to Trajectory Sampling Schemes
209 In our study, we sample locations of vehicle 𝑖𝑖 every 1 second, but the original data sets provide
210 the locations every one tenth or one fifteenth second. Thus the starting point 𝑡𝑡 𝑖𝑖 ( 0) could have an
211 impact on speeds, accelerations, and total travel distance. Table 1 shows the mean and standard
212 deviation of different traffic and emission values for four vehicles from the four data sets. Here
213 we assume that all vehicles are of LDV3 type. From the table we can see that the coefficient of
214 variation is generally smaller than 2%. Thus emission estimation is not sensitive to the sampling
215 scheme.
216
8
Dataset ID 𝑇𝑇 𝑖𝑖
( s) 𝑑𝑑 𝑖𝑖
( km) 𝑣𝑣 𝑖𝑖
( km/ h)
HC
( mg/ km)
CO
( mg/ km)
NOX
( mg/ km)
CO2
( g/ km)
Fuel
( l/ km)
Mean
( Std)
2: 35-
3: 05
3284 50.6000
( 0.5071)
0.8678
( 0.0097)
61.7423
( 0.2479)
16.2365
( 0.1917)
349.5191
( 5.7651)
74.5857
( 2.8629)
119.50
( 1.44)
0.0515
( 0.0006)
Mean
( Std)
4: 00-
4: 15
672 29 ( 0) 0.4691
( 0.0039)
58.2290
( 0.4806)
47.7081
( 0.9020)
1281.1
( 45.7)
151.0042
( 5.6914)
312.21
( 8.37)
0.1371
( 0.0034)
Mean
( Std)
5: 00-
5: 15
284 51.6000
( 0.5164)
0.4702
( 0.0056)
32.8060
( 0.0887)
21.1447
( 1.0149)
498.0905
( 37.1156)
71.9498
( 2.1348)
227.58
( 7.23)
0.0978
( 0.0033)
Mean
( Std)
5: 15-
5: 30
3 114.3000
( 0.4830)
0.4766
( 0.0025)
15.0109
( 0.0306)
24.6502
( 0.0722)
542.8825
( 1.8056)
62.9381
( 0.3373)
339.25
( 0.42)
0.1453
( 0.0002)
Table 1. Sensitivity of emissions and fuel consumption 217 to trajectory sampling scheme
218 3.2 Impacts of speed- acceleration adjustment
219 As discussed in Section 2, the VT- Micro model only works for certain speed- acceleration
220 domain. Also we showed that many points from the data set of 2: 35- 3: 05 are outside of the
221 domain. In Figure 3, we demonstrate possible impacts of speed- acceleration adjustment. Here we
222 assume that all vehicles are of LDV3 type. In both Figure 3( a) and 3( b), the dots are CO2
223 emissions after green driving, and the cusp at 120 km/ h in Figure 3( a) is caused by speed
224 adjustment and can be explained with Equation 8.
225
226 Figure 3. Impacts of speed- acceleration adjustment on CO2 emissions
227 In Figure 3, the asterisks are CO2 emissions before green driving. In Figure 3( b), emission
228 values greater than 600 g/ km are set to 600 g/ km. Comparing both figures we can see that CO2
229 emissions at high speeds can be substantially impacted by speed- acceleration adjustment. In this
230 sense, the fidelity of the VT- Micro model is low for traffic near free flows. Therefore, in later
231 studies, we will choose vehicles for which the difference between the emissions calculated with
232 speed- acceleration adjustment and those without adjustment is within 5%. Namely, we will
233 choose vehicle i if for any 𝑒𝑒 = 1,⋯, 5 234 𝐸𝐸 𝐸𝐸 𝑒𝑒 𝑒𝑒 ,, 𝑖𝑖 𝑖𝑖 (( 𝒗𝒗 𝒗𝒗 􀷝 𝑖𝑖 𝑖𝑖 ,, 𝒂𝒂 𝒂𝒂 􀷝 𝑖𝑖 𝑖𝑖 ))∈[ 0.95,1.05] ( 10)
235 With this criterion, we are able to choose 105 out of 4543 cars in the 2: 35- 3: 05 data set, 1752 out
236 of 1942 cars in the 4: 00- 4: 15 data set, 1622 out of 1742 cars in the 5: 00- 5: 15 data set, and 1648
9
out 1724 cars in the 5: 15- 5: 30 data set. That is, 237 more vehicles' speeds and acceleration rates are
238 within the VT- Micro's feasible domain when traffic gets more congested.
239 4 ENVIRONMENTAL BENEFITS OF GREEN DRIVING STRATEGIES FOR
240 DIFFERENT CONGESTION LEVELS
241 In this section, we analyze potential savings of green driving strategies for 5127 vehicles
242 satisfying Equation 10. Hereafter, we use original speeds and acceleration rates without
243 adjustment in Equations 7 and 8.
244 4.1 Potential Savings of Green Driving Strategies
245 Figure 4 shows emissions of four pollutants and fuel consumptions of 5127 vehicles. These
246 vehicles' total travel time is 116.96 hours, and the total travel distance is 2448.2 km. Before
247 green driving, the total amount of HC, CO, NOX, CO2, and fuel consumption for all vehicles are
248 0.0557 kg, 1.2656 kg, 0.1764 kg, 682.5 kg, and 292.5 l, respectively. After green driving, the
249 total amount are 0.0470 kg, 1.1160 kg, 0.1224 kg, 544.7 kg, and 234.4 l, respectively. The
250 corresponding savings in total emissions and fuel consumptions are 15.63%, 11.81%, 30.58%,
251 20.19%, and 19.88%. Furthermore, we can see that the average gas mileage increases from 11.95
252 l/ 100 km or 19.69 mile/ gallon to 9.57 l/ 100 km or 24.57 mile/ gallon.
253
254 Figure 4. Emissions and fuel consumptions of vehicles before ( blue asterisks) and after ( red
255 dots) green driving strategies
10
Figure 5 shows potential savings with green driving strategies. F 256 rom both Figures 4 and 5, we
257 can see that it is possible that the proposed green driving strategies could increase emissions and
258 fuel consumptions. In particular, 0.86%, 31.23%, 0.23%, and 0.66% of all vehicles could
259 increase their emissions in HC, CO, NOX, and CO2, respectively, and 0.66% could increase
260 their fuel consumption. The ranges of average speeds for the occurrence of such increases are [ 10,
261 40] km/ h for HC, [ 5, 40] km/ h for CO, [ 35, 90] km/ h for NOX, [ 35, 90] km/ h for CO2, and [ 35,
262 90] km/ h for fuel consumption. That is, generally green driving strategies could increase HC and
263 CO for speeds lower than 40 km/ h and NOX, CO2, and fuel consumption for speeds higher than
264 35 km/ h.
265
266
267 Figure 5. Potential savings in emissions and fuel consumptions with green driving strategies
268 Even though a significant portion of vehicles could increase their CO emissions, but from Figure
269 5( b) we can see that other vehicles can decrease their CO emissions by a even larger margin.
270 Figure 6 further shows the distribution patterns of saving percentages. This figure confirms our
271 observations from Figures 4 and 5. From the figure we can also see that CO2 and fuel have
272 almost the same saving profile. This means that they are highly related to each other.
11
273
274 Figure 6. Distribution of saving percentages for different emissions and fuel consumption
275 4.2 Fitting the relationships between emissions/ fuel consumptions and
276 average speeds
277 As can be seen from Figures 4 and 5, we are only able to estimate potential savings for vehicles
278 with average speeds smaller than 90 km/ h. In this subsection, we attempt to fit emissions and
279 fuel consumptions as functions of average speed and then extrapolate the resulted curves to free
280 flow regime. Here we adopt the same functions from reference ( 6) for emissions and fuel
281 consumptions of vehicle i before green driving 282 ln 􀵫 𝐸𝐸 𝑒𝑒 , 𝑖𝑖 ( 𝒗𝒗 𝑖𝑖 , 𝒂𝒂 𝑖𝑖 ) 􀵯 = 𝛽𝛽 𝑒𝑒 ,0+ 𝛽𝛽 𝑒𝑒 ,1 𝑣𝑣 𝑖𝑖 + 𝛽𝛽 𝑒𝑒 ,2 𝑣𝑣 𝑖𝑖 2+ 𝛽𝛽 𝑒𝑒 ,3 𝑣𝑣 𝑖𝑖 3+ 𝛽𝛽 𝑒𝑒 ,4 𝑣𝑣 𝑖𝑖 4+ 𝜖𝜖 𝑒𝑒 , 𝑖𝑖 ( 11)
283 where 𝐸𝐸 𝑒𝑒 , 𝑖𝑖 ( 𝒗𝒗 𝑖𝑖 , 𝒂𝒂 𝑖𝑖 ) is defined in Equation 7 with second- by- second speeds 𝒗𝒗 𝑖𝑖 and acceleration
284 rates 𝒂𝒂 𝑖𝑖 , 𝑣𝑣 𝑖𝑖 is the average speed of vehicle i, 𝛽𝛽 𝑒𝑒 , 𝑗𝑗 ( 𝑒𝑒 = 1,⋯, 5; 𝑗𝑗 = 0,⋯, 4) is the coefficient for
285 emission or fuel e, and 𝜖𝜖 𝑒𝑒 , 𝑖𝑖 is the error term. The same formula can also be used to fit emissions
286 and fuel consumptions after green driving.
Pollutants 𝛽𝛽 𝑒𝑒 ,0 𝛽𝛽 𝑒𝑒 ,1 𝛽𝛽 𝑒𝑒 ,2 𝛽𝛽 𝑒𝑒 ,3 𝛽𝛽 𝑒𝑒 ,4 𝑅𝑅 2
e= 1
( HC)
Before 4.027244 - 0.08204 0.002159 - 2.1E- 05 6.13E- 08 0.6189
After 4.011591 - 0.09116 0.002683 - 3.9E- 05 2.06E- 07 0.9978
e= 2
( CO)
Before 7.111451 - 0.08096 0.002174 - 2.1E- 05 5.84E- 08 0.5110
After 7.069401 - 0.0757 0.002208 - 3.5E- 05 2.06E- 07 0.9985
e= 3
( NOX)
Before 4.703609 - 0.06186 0.002251 - 2.7E- 05 1.11E- 07 0.5958
After 4.753301 - 0.09837 0.003334 - 4.3E- 05 2.06E- 07 0.9947
e= 4
( CO2)
Before 13.7517 - 0.1002 0.002564 - 2.9E- 05 1.17E- 07 0.8777
After 13.75014 - 0.12122 0.003287 - 4.2E- 05 2.06E- 07 0.9988
e= 5
( Fuel)
Before - 0.91003 - 0.1003 0.002556 - 2.9E- 05 1.13E- 07 0.8750
After - 0.91634 - 0.12052 0.003271 - 4.2E- 05 2.06E- 07 0.9988
287 Table 2. Results for fitting the relationships between emissions/ fuel consumptions and
288 average speeds with the model in Equation 11
289 Table 2 shows the coefficients and R- square for the relationships between four emissions and
290 fuel consumptions and average speeds. From the table, we can see that, after green driving, the
12
model in Equation 11 can fit the data very well with R- square> 291 0.99. This is not coincidence.
292 Rather, from Equations 1 and 8, we can see that 293 ln 𝐸𝐸 􀴤 𝑒𝑒 , 𝑖𝑖 =− ln36 𝑣𝑣 0 𝑖𝑖 0− ln 𝑀𝑀 𝑀𝑀 𝐸𝐸 𝑒𝑒 ( 𝑣𝑣 𝑖𝑖 ,0)=− ln36 𝑣𝑣 0 𝑖𝑖 0− 𝐿𝐿 𝑒𝑒 0,0− 𝐿𝐿 𝑒𝑒 1,0 𝑣𝑣 𝑖𝑖 − 𝐿𝐿 𝑒𝑒 2,0 𝑣𝑣 𝑖𝑖 2− 𝐿𝐿 𝑒𝑒 3,0 𝑣𝑣 𝑖𝑖 3 , ( 12)
294 which can be well approximated with Equation 11. For emissions and fuel consumptions before
295 green driving, the model in Equation 11 can fit the data well with R- square> 0.5. In particular, the
296 model fits CO2 and fuel consumptions with R- squares greater than 0.87. However, the model in
297 Equation 11 cannot be used to extrapolate data to free- flow regime well. This can be seen in
298 Figure 7, which shows that emissions are unreasonably large for large average speeds after green
299 driving.
300
301 Figure 7. Fitted relationships between emissions/ fuel consumptions and average speeds
302 with Equation 11: dashed lines for relationships before green driving, solid lines for those
303 after green driving.
Pollutants 𝛽𝛽 𝑒𝑒 ,0 𝛽𝛽 𝑒𝑒 ,1 𝛽𝛽 𝑒𝑒 ,2 𝛽𝛽 𝑒𝑒 ,3 𝑅𝑅 2
e= 1
( HC)
Before - 2.84072 0.037989 - 5.7E- 05 - 9E- 07 0.9429
After - 3.01905 0.05401 - 0.00077 4.66E- 06 1
e= 2
( CO)
Before 0.246721 0.038572 - 1.8E- 05 - 1.3E- 06 0.9322
After 0.038755 0.06947 - 0.00124 7.7E- 06 1
e= 3
( NOX)
Before - 2.22054 0.066847 - 0.00039 7.48E- 07 0.9654
After - 2.27734 0.0468 - 0.00011 2.23E- 07 1
e= 4
( CO2)
Before 6.821611 0.029429 - 0.00012 - 4.1E- 07 0.8975
After 6.719495 0.02395 - 0.00016 8.36E- 07 1
e= 5
( Fuel)
Before - 7.83607 0.028707 - 0.0001 - 5.7E- 07 0.8980
After - 7.94699 0.02465 - 0.00018 9.31E- 07 1
304 Table 3. Results for fitting the relationships between emissions/ fuel consumptions and
305 average speeds with a new model in Equation 13
13
Inspired by Equation 12, we introduce a new model 306 for the relationships: 307 ln 􀵫 𝐸𝐸 𝑒𝑒 , 𝑖𝑖 ( 𝒗𝒗 𝑖𝑖 , 𝒂𝒂 𝑖𝑖 ) 􀵯 + ln36 𝑣𝑣 0 𝑖𝑖 = 𝛽𝛽 𝑒𝑒 ,0+ 𝛽𝛽 𝑒𝑒 ,1 𝑣𝑣 𝑖𝑖 + 𝛽𝛽 𝑒𝑒 ,2 𝑣𝑣 𝑖𝑖 2+ 𝛽𝛽 𝑒𝑒 ,3 𝑣𝑣 𝑖𝑖 3+ 𝜖𝜖 𝑒𝑒 , 𝑖𝑖 . ( 13)
308 Table 3 shows the coefficients and R- square for the relationships between four emissions and
309 fuel consumptions and average speeds. Comparing results in Table 3 with those in Table 2, we
310 can see that the new model in Equation 13 has one fewer parameter than that in Equation 11, but
311 it fits the data very well: it is not surprising that R- squares are 1 after green driving as explained
312 in Equation 12; even for data before green driving, R- squares are close to or larger than 0.9 for
313 all the variables.
314 Figure 8 shows fitted relationships between emissions/ fuel consumptions and average speeds.
315 Compared with Figure 7, these curves look more reasonable. Therefore, we could apply Equation
316 13 to extrapolate for emissions and fuel consumptions in free flow regime. From Figure 8, we
317 can see that, on average, green driving strategies may not be able help to save emissions when
318 average speeds are greater than 90 km/ h. However, it is important to note that such a conclusion
319 is based on the assumption that Equation 13 can be extrapolated to free flow regime.
320
321 Figure 8. Fitted relationships between emissions/ fuel consumptions and average speeds
322 with Equation 13: dashed lines for relationships before green driving, solid lines for those
323 after green driving.
324 Figure 9 shows saving percentages with fitted relationships in Equation 13. In the figure, all
325 negative percentages are set to be 0. As shown in the figure, the average speeds with most
326 savings are around 70 km/ h for HC and CO, 40 km/ h for NOX, and 50 km/ h for CO2 and fuel
327 consumption. Furthermore, the saving percentages can be greater than 40% for HC, 60% for CO,
328 30% for NOX, and 20% for CO2 and fuel consumption. Note that these observations are
329 probably true, since Equation 13 can fit data before and green driving very well for average
330 speeds smaller than 90 km/ h.
14
331
332 Figure 9. Saving percentages from fitted relationships between emissions/ fuel consumptions
333 and average speeds with Equation 13
334 5 DISCUSSION AND CONCLUSIONS
335 In this study with the help of NGSIM data sets and the VT- Micro model we attempted to provide
336 some insights on potential environmental benefits of green driving strategies. We carefully
337 discussed the sampling and calculation of vehicles' second- by- second speeds and acceleration
338 rates and examined them against VT- Micro's feasible domain of speed and acceleration. We
339 found that, when average speeds are higher than 90 km/ h, vehicles' second- by- second speeds and
340 acceleration rates have to be adjusted. As a result of such an adjustment, vehicle emissions and
341 fuel consumptions can be significantly different from the original ones. Therefore we only chose
342 5127 out of 9951 cars in the original data sets for our further study. With the chosen vehicles,
343 we discussed potential savings in fuel consumptions and emissions of HC, CO, NOX, and CO2.
344 In addition, we presented a new model, Equation 13, to fit the relationships between
345 emissions/ fuel consumptions and average speeds.
346 In this study, we made the following contributions. First, we presented a systematic procedure to
347 sample and calculate vehicles' second- by- second speeds and acceleration rates and estimate
348 vehicle emissions from NGSIM data. We also demonstrated impacts of speed- acceleration
349 adjustment when applying the VT- Micro model and presented a criterion to choose reasonable
350 vehicle trajectories. This procedure is critical for obtaining meaning results and insights. Second,
351 we proposed a new model, Equation 13, for the relationships between emissions/ fuel
352 consumptions and average speeds. Compared with an existing model in ( 6), this model has one
353 fewer parameter, but provides significantly better fitting with R- squares near or greater than 0.9
354 for all five variables. The new model also provides more reasonable results for speeds higher
355 than 90 km/ h. Third, we found that, from Figure 9, the saving profiles are usually concave with
15
respect to speeds with a unique maximum points. In addition, 356 we found that the saving
357 percentages can be greater than 40% for HC, 60% for CO, 30% for NOX, and 20% for CO2 and
358 fuel consumption. Such savings are usually achieved for speeds around 50 km/ h. We have a high
359 confidence in these observations since ( 1) we used 5127 vehicles' trajectories and ( 2) the new
360 model in Equation 13 fits the data with very high R- squares. The most important insight from
361 this study is that green driving strategies that can help to smooth traffic flow can achieve the best
362 effects for traffic flow with an average speed around 50 km/ h.
363 However, this study has a number of limitations. The first limitation is in the assumption of
364 constant driving speeds for all vehicles after using green driving strategies. In reality, this can
365 hardly be achieved simultaneously. However, in our other study ( 12), it is shown that smooth
366 driving can indeed be achieved with the help of inter- vehicle communications for small market
367 penetration rates and reasonable communication delays. The second limitation is related to the
368 VT- Micro model, which only works for relatively congested traffic. In the future, we will check
369 out other models such as MOVES and CMEM. The third limitation is related to the NGSIM data,
370 from which we cannot tell vehicle model year, engine size, or mileage. Thus we cannot
371 determine the exact vehicle type used in VT- Micro or other emission models. In this study, we
372 simply assume all vehicles are of LDV3 type. That said, however, we suspect that different
373 vehicle types will not significantly impact the relative savings.
374 In addition to addressing the aforementioned limitations of this study, we will be interested in
375 investigating potential environmental benefits for more traffic patterns with other NGSIM data
376 sets. We will also investigate how to implement green driving strategies with information and
377 communication technologies, especially IntelliDrive technologies. We will also test such green
378 driving strategies in field in the follow- up studies.
379 ACKNOWLEDGEMENTS
380 This study is supported in part by a grant from the University of California Transportation Center.
381 We would also like thank Dr. Jean- Daniel Saphores for his discussions. The views and results
382 are the authors' alone.
383
384 References
385 ( 1) Greene, D. L., and Schafer, A. ( 2003). “ Reducing Greenhouse Gas Emissions from U. S.
386 Transportation, Prepared for the Pew Center on Global Climate Change.” Available from
387 http:// www. pewclimate. org/ docUploads/ ustransp. pdf.
388 ( 2) Bose, A., Ioannou, P., Inc, R., and Sunnyvale, C. ( 2003). Analysis of traffic flow with mixed
389 manual and semiautomated vehicles. Intelligent Transportation Systems, IEEE Transactions on,
390 4( 4): pp. 173– 188.
391 ( 3) K. Ahn and H. Rakha. ( 2008). “ The effects of route choice decisions on vehicle energy
392 consumption and emissions.” Transportation Research Part D, 13( 3): pp. 151– 167.
393 ( 4) Barth, M. and Boriboonsomsin, K. ( 2009). “ Energy and emissions impacts of a freeway-394
based dynamic eco- driving system.” Transportation Research Part D, volume 14, ( 6), pp. 400—
395 410.
16
( 5) Ihab El- Shawarby, Kyoungho Ahn, Hesham Rakha, ( 2005). “ Comparative 396 field evaluation of
397 vehicle cruise speed and acceleration level impacts on hot stabilized emissions” Transportation
398 Research Part D 10, pp. 13– 30
399 ( 6) Barth, M. and Boriboonsomsin, K. ( 2008). “ Real- World Carbon Dioxide Impacts of Traffic
400 Congestion” Transportation Research Record No. 2058, Transportation Research Board of the
401 National Academies, Washington, D. C., pp. 163– 171.
402 ( 7) M. Treiber, A. Kesting, C. Thiemann. ( 2008). “ How Much does Traffic Congestion Increase
403 Fuel Consumption and Emissions? Applying a Fuel Consumption Model to the NGSIM
404 Trajectory Data” TRB paper 2008
405 ( 8) K. Ahn, H. Rakha, A. Trani and M. Van Aerde, ( 2002) “ Estimating vehicle fuel- consumption
406 and emissions based on instantaneous speed and acceleration levels” Journal of Transportation
407 Engineering 128, pp. 182– 190
408 ( 9) Int Panis, L. and Broekx, S. and Liu, R. ( 2006) “ Modelling instantaneous traffic emission and
409 the influence of traffic speed limits”. Science of the Total Environment, volume 371, ( 1- 3), pp.
410 270— 285.
411 ( 10) Liu, R. and Tate, J. ( 2004). “ Network effects of intelligent speed adaptation systems”,
412 Transportation, volume 31, ( 3), pp. 297-- 325.
413 ( 11) Box, G. E. P. and Cox, D. R. ( 1964). “ An analysis of transformations”. Journal of the Royal
414 Statistical Society. Series B ( Methodological), volume ( 26), ( 2), pp. 211-- 252.
415 ( 12) Yang, H., D. Yuan, W.- L. Jin and J.- D. Saphores ( 2010). " Simulation evaluation of green
416 driving strategies based on inter- vehicle communications". Submitted to TRB 2011 Annual
417 Meeting.