Beruflich Dokumente
Kultur Dokumente
Abstract: This paper addresses the question of what speed profile will minimize fuel consumption of a land transport vehicle 共road or
rail兲 in traversing a path or route. Numerous previous studies, using a control theory approach applied to specific profiles, have suggested
that fuel consumption is approximately minimized by operation at constant speed. This result is derived much more directly here, along
with boundaries on the conditions under which this result holds. The derivation relies on: 共1兲 the approximate proportionality between fuel
consumption and propulsive work; 共2兲 the inherent resistance of a vehicle having the usual quadratic form 共of road and rail vehicles兲; 共3兲
the energy conversion characteristics of on-board gasoline or diesel 共or diesel–electric兲 propulsion systems; and 共4兲 relatively long
distances between stops. Tests, using a train performance simulator, confirm the theoretical results. The results are discussed from the
standpoint of basic principles in transportation engineering, and as guidance for designing systems that conserve fuel.
DOI: 10.1061/共ASCE兲0733-947X共2005兲131:3共173兲
CE Database subject headings: Fuel consumption; Optimization; Rail transportation; Highway transportation; Road design; Speed
control; Transportation engineering.
Introduction train performance and fuel consumption. The test results are con-
sistent with the theoretical results and also provide an indication
This paper addresses the question of what speed profile will mini- of the magnitude of fuel consumption increase resulting from
mize propulsive work expended by a land transport vehicle 共road deviations from constant speed. The paper concludes with a dis-
or rail兲 in traversing a path. The results of specific applications of cussion of the results and their significance, both as an interesting
numerous previous studies of optimal speed profiles have sug- theoretical result in transportation engineering and as guidance
gested that under a wide range of path conditions 共line profile, for designing systems that conserve fuel.
curvature, speed limits兲 that fuel consumption of a motor vehicle
or train is approximately minimized by operation at constant Background
speed 共e.g., Schwarzkopf and Leipnik 1977兲. It is shown here that
under quite general conditions this is true. To derive general re- Before delving into the analysis and results, it is important to
sults, in this effort the fact that fuel consumption has been shown discuss the reason this research was undertaken. At the most gen-
to be approximately proportional to propulsive work is used. This eral level, minimizing fuel consumption is an important compo-
enables derivation of the optimal speed profile with respect to nent of efficiency. It is also important from the standpoint of
minimizing propulsive work, and this will minimize fuel con- conserving natural resources and reducing pollutant emissions.
sumption assuming proportionality between work and fuel con- The advent of various advances in vehicle control systems has
sumed. The key additional assumptions are that the vehicle’s in- prompted considerable interest in using such technology to not
herent vehicle resistance follows the usual quadratic form used to only increase capacity but also to improve the efficiency of op-
describe road and rail vehicles, that the usual energy conversion eration 共Smith et al. 1990兲. One aspect of operating efficiency is
limitations of on-board gasoline or diesel 共or diesel–electric兲 pro- of course fuel or energy conservation, which is important not only
pulsion systems apply, and that distances traversed without stops because of cost and resource use but also because of pollution
are relatively long. The rules are then tested using a simulator of reduction. Much work has been done using control theory to de-
termine fuel-minimizing speeds and throttle settings. However,
1
Research Fellow, Transportation and Logistics Systems Laboratory, these methods require detailed analysis of both the vehicle and
Univ. of Pennsylvania, 220 S. 33rd St., Philadelphia, PA 19104-6315. the guideway path, and the results are unique to each. In this
E-mail: davidchang@earthlink.net paper we try to develop general guidelines that do not require
2
UPS Foundation Professor of Transportation and Professor of such a detailed analysis. The guidelines developed here are basi-
Systems Engineering, Univ. of Pennsylvania, 220 S. 33rd St., Room 229, cally consistent with results that have been reported in the control
Towne Building, Philadelphia, PA 19104-6315 共corresponding author兲. theory literature but are more complete in identifying boundaries
E-mail: morlok@seas.upenn.edu of their validity and are much simpler to apply. These should be
Note. Discussion open until August 1, 2005. Separate discussions useful in systems that have the capability to pace vehicles over
must be submitted for individual papers. To extend the closing date by
long distances.
one month, a written request must be filed with the ASCE Managing
Editor. The manuscript for this paper was submitted for review and pos-
sible publication on July 25, 2003; approved on April 20, 2004. This Approach and Assumptions
paper is part of the Journal of Transportation Engineering, Vol. 131,
No. 3, March 1, 2005. ©ASCE, ISSN 0733-947X/2005/3-173–182/ The basic approach used in this research is deductive. The deri-
$25.00. vation begins with an analysis of the most basic case of a vehicle
冕
sions in Table 1兲 are given in Table 2.
x2
Railroad trains normally consist of multiple units 共one or more
W= Pm共x兲dx 共5兲
locomotive units and cars兲. The total inherent resistance 共R兲 of the x1
entire vehicle 共train兲 is the sum of the inherent resistance of each
locomotive or car in the train 共RI兲. These are given by the follow- This basic equation appears in various forms in the rail and
ing equations; road literature. For example, for automobiles the Automotive
handbook 共Bosch 1996, p. 554兲 presents an equation for propul-
R I = a I + b Iv + c Iv 2 共2兲 sive force 共termed there “thrust and tractive forces”兲 that has the
inherent resistance on a level road as the following function of Most of the prior work relates to the speed of road vehicles using
speed: essentially control theory approaches. The inputs to such models
R = TU + CA共/2兲v2 共6兲 are detailed descriptions of the road or rail line, including gradi-
ents, speed limits, and in the case of some models for railroads,
The various constants 共all terms except v and R兲 are defined in curves. The output identifies the points at which either control
Table 1. settings of the power unit 共throttles, brakes, etc.兲 or desired speed
The total resistance equals the sum of the inherent resistance are to be changed, and the new setting. In one of the earliest
共R兲, any added curvature resistance 共H兲, if the vehicle is rounding works, Schwarzkopf and Leipnik 共1977, pp. 279–286兲 developed
a curve, and any gradient force F, if the vehicle is ascending or a nonlinear optimal control model of motor vehicle throttle set-
descending a grade. Table 4 presents typical gradient and curva- tings to minimize fuel consumption under varied road conditions.
ture resistance expressions for automobiles, trucks, buses, and Schwarzkopf and Leipnik concluded that the optimal velocity for
trains. Note for motor vehicles H is taken as zero 共probably be- a level road is a steady state velocity. The optimal velocity for
cause the added resistance is nil or very small兲. For trains, H is various grades was not found in closed form, but is to be found
simply a function of vehicle weight and the degree of curvature through use of a heuristic procedure. Hooker 共1988, pp. 183–
共or equivalently, radius of curvature兲. The fact that neither gradi- 201兲, building on Schwarzkopf and Leipnik, developed a model
ent nor curvature resistance is a function of speed is important for for minimizing motor vehicle fuel consumption subject to speed
the later derivations. and time constraints over a specified distance. It also considers
Thus we have the required propulsion force P equal to the gradients. He applied his method to a variety of model vehicles,
total resistance of the vehicle plus the acceleration force, if any and found that constant speed satisfies the optimality conditions
P = max共E,0兲 共7a兲 for a constant road grade. In fact such a result is valid for only a
certain range of gradients, as will be derived later in this paper.
Numerous others, including Kormanski et al. 共1983兲, developed
E = R + H + F + M v̇ 共7b兲
similar models and to varying degrees had results that suggested
where M⫽mass of the vehicle; v̇⫽its acceleration; and at least close to constant speed operation in order to minimize fuel
P⫽propulsion force. consumption.
Note that F can be positive 共for an upgrade兲 or negative 共for a In the rail transportation area, the Scheduling and Control
downgrade兲. If the resultant force E is less than zero, then no Group at the Univ. of South Australia have developed an exten-
propulsion force is needed. A braking force is applied to control sive set of models for train operation based on optimal control
vehicle speed as necessary. In some applications, particularly au- theory. One result was that where operation at constant speed is
tomobiles, the rolling resistance portion of inherent resistance is possible on level track, it is optimal 共Howlett and Pudney 1995, p.
adjusted downward on a nonzero gradient 共as implied in Table 1兲, 30兲. In a more empirical vein, a major study at the Transportation
to reflect the reduced load on the bearings and wheels. Of signifi- Systems Center in 1975–1982 共Hopkins et al. 1975, 1977, 1978兲
cance is that this adjustment is not a function of speed. published fuel consumption measurements and estimates for vari-
The important conclusion from the foregoing is that road and ous rail freight service scenarios, including variations in train
rail vehicles have the same general form for the total resistance speed, size 共weight and length兲, power to weight ratio, and terrain
equation and the same form for the propulsive work equation. The 共path profile兲. While they did not address the optimal speed ques-
coefficients a, b, and c in the inherent resistance portion, and the tion directly, Hopkins noted that, in scenarios examined, a non-
constants in the curvature resistance portion differ only by their uniform velocity profile could consume 5–15% more fuel than
magnitude between the various land based vehicles. The gradient would a constant speed profile, while yielding the same average
force term is identical for all vehicles, of course. This provides speed 共Hopkins 1982兲. Also of note is the fact that the basic
the basis for the analytical treatment to follow that yields optimal model used in the analytical portions of the effort was essentially
speed results that will apply similarly to various vehicles. a propulsive work relationship.
Two assumptions should be mentioned. In the analysis below, Thus prior works in both the road and rail modes suggest that
it is assumed that the inherent resistance equation parameters constant speed operation is a good policy from the standpoint of
共a, b, and c兲 of the vehicle remain unchanged as the vehicle minimizing fuel consumption, at least on a level guideway. How-
traverses its path. In particular, this means that the reduction in ever, the analytical results depend upon various relatively com-
vehicle weight due to any consumption of fuel is negligible and plex methods for arriving at the conclusion. Also, all of these are
can be ignored. Also, the approximation that fuel consumption is intended to apply to specific cases in which detailed vehicle char-
冋冕 册
not change the result. This result can be shown by examining the
speed profile with acceleration 共and deceleration兲 included in the x=B
determination of propulsive work 共and hence fuel consumption兲. zA−B = r R共v共x兲兲 · dx + 0.5M共vL兲2 − 0.5M共vA兲2 共34兲
Fig. 1 shows the speed profile of a vehicle that has an overall x=A
average speed of vA but that operates at a lower speed 共vL ⬍ vA兲 in The integral term can be replaced by the product of the average
one section 共A – D兲 and a higher speed vH 共vH ⬎ vA兲 in the other inherent resistance of the vehicle as it decelerates from vA at
section 共D – G兲. The beginning and ending speeds are identical location A to vL at location B. Thus we can write
and equal vA. The vehicle speed at location A is vA. The vehicle
speed is reduced, and at B the lower speed vL is achieved. This zA–B = r关共RA–B * SA–B兲 + 0.5 * M * 共vL兲2 − 0.5 * M * 共vA兲2兴 共35兲
speed is maintained to C. At C the vehicle starts to increase speed where
冋冕 册
to vA, and vA is reached at D. The speed continues to increase to
x=B
the higher speed vH, which is reached at E. This speed is main-
tained to F, at which point the vehicle starts to decrease speed to RA–B = R共v共x兲兲 · dx /SA–B 共36兲
x=A
vA, which is reached at location G. This speed profile will be
referred to as the actual speed profile. Thus RA–B⫽average inherent resistance of the vehicle as it decel-
Let the fuel consumption of the vehicle with the actual speed erates over segment A – B. It is useful to note at this point that
profile from location A to location G be zA–G. This is the sum of corresponding to RA–B is an equivalent speed vA–B that has an
the fuel consumed in each section inherent resistance equal to RA–B.
Similarly the actual propulsive work performed in the section
zA–G = zA–B + zB–C + zC–D + zD–E + zE–F + zF–G 共30兲 C – D is the propulsive work to overcome the inherent resistance
where zi–f ⫽fuel consumption in section i – f. From Eq. 共4兲 the fuel and the additional work expended to increase vehicle speed.
consumption zB–C in section B – C is the product of the fuel rate r, Again using an average inherent resistance RC–D, the resulting
the inherent resistance RL at speed vL, and the length of the sec- fuel consumption is
tion SB–C
zC–D = r关共RC–D * SC–D兲 + 0.5 * M * 共vA兲2 − 0.5 * M * 共vL兲2兴 共37兲
zB–C = r共RL兲SB–C 共31兲 Note that in Eqs. 共35兲 and 共37兲 the terms for change in kinetic
Similarly energy are identical but of opposite sign. Thus these two terms
cancel one another when these are substituted into Eq. 共30兲. This
zE−F = r共RH兲SE−F 共32兲 makes intuitive sense: The propulsive work saved when decreas-
However, the fuel consumption in each of the other sections ing speed from vA to vL equals the additional propulsive work
共zA−B, zC−D, zD−E, and zF−G兲 includes both overcoming inherent required when increasing speed from vL to vA, so these two terms
resistance and acceleration or deceleration. The net change in cancel out, leaving just the propulsive work to overcome the in-
propulsive work due to the changing of speed equals the change herent resistance.
in kinetic energy 共Serway 1996兲 The same cancellation occurs with the kinetic energy terms for
sections D – E and F – G. Thus the total fuel consumption over the
Wi−f = K f − Ki = 0.5M v f 2 − 0.5M vi2 共33兲 distance A – G is
where K⫽kinetic energy; K f ⫽kinetic energy at location f; zA–G = r关共RA–B * SA–B兲 + 共RL * SB–C兲 + 共RC–D * SC–D兲
Ki⫽kinetic energy at location i; M⫽mass of vehicle; v f ⫽speed of
+共RD–E * SD–E兲 + 共RH * SE–F兲 + 共RF–G * SF–G兲兴 共38兲
vehicle at location f; vi⫽speed of vehicle at location i; and
Wi−f ⫽work performed or saved due to changing speed from lo- As observed earlier, RA–B has a corresponding speed that is
cation i to f. between vL and vA. Similarly the average resistance yields resis-
Thus the actual propulsive work performed in segment A – B tance in each of the segments C – D , D – E, and F – G and has a
equals the propulsive force to overcome the inherent resistance corresponding speed lying between the initial and final speeds in
and the propulsive work saved by the vehicle reducing speed that segment. The relative magnitude of these speeds is shown in
from vA to vL. This assumes that all kinetic energy saved is used Fig. 2. This profile with equivalent speeds is called the
and the intermodal freight train used earlier. The first consists of substantial speed variation 共and the highest standard deviation of
operating a train on a level tangent track divided into two equal speed among the three scenarios兲.
sections, each 16 km 共10 mi兲 long, following the velocity profile The standard deviation 共SD兲 of vehicle speed is plotted against
of Fig. 1 共and the inverse of a higher speed in the first section兲. fuel consumption for each scenario in Fig. 5. The standard devia-
The speeds were chosen so as to achieve the intended overall tion was calculated from the speeds recorded by the TPS program
average speed of 64 km/ h 共40 mi/ h兲 in each case. One hundred at the 163 markers on the line 共at points of change in gradient or
runs were made with different combinations of speeds. curves兲. As expected, Scenario 1 had the lowest speed variation,
The results are shown in Fig. 4, with the ratio of the speeds on with Scenarios 2 and 3 successively higher. Fuel consumption
the two sections 共v2 / v1兲 on a log scale on the horizontal axis, and followed a similar pattern. Scenario 1 had a SD of 17.4 km/ h
the fuel consumption per unit distance in liters per kilometer 共10.8 mi/ h兲, and fuel consumption of 4,159.8 L 共1,097.6 gal兲.
共L / km兲 on the vertical axis. The figure shows that fuel consump- Scenario 3 had a SD of speed of 26.7 km/ h 共16.6 mi/ h兲, a 53.4%
tion is a minimum when the speed ratio is unity and that fuel increase, with fuel consumption of 4,645.5 L 共1,225.7 gal兲, an
consumption increases rapidly for deviations from the uniform 11.7% increase.
speed condition. These results are consistent with the analytic These results are consistent with field measurements of rail
result that fuel is minimized with a uniform speed. The log scale fuel consumption made in the previously mentioned USDOT rail
is used in this figure to make the figure visually symmetric. With energy study. Hopkins et al. 共1978, p. 60兲 reported finding that a
a uniform speed of 64 km/ h 共40 mi/ h兲, the fuel consumption is nonuniform velocity profile can result in consuming from 5 to
10.02 L / km 共4.26 gal/ mi兲. With a speed ratio 共v2 / v1兲 equal to 15% more fuel than would be the consumed at a constant speed
1.5 关average speed of 88.6 km/ h 共55 mi/ h兲 over the first section, 共with the same average speed兲.
followed by about 58.5 km/ h 共37 mi/ h兲 in the second section兴, Thus the theoretical result that a constant speed is the optimal
the fuel consumption increases by 60% to 16 L / km speed profile under a wide range of conditions is supported by
共6.83 gal/ mi兲. Thus varying speed can result in a substantial in- these simulation results.
crease in fuel consumption.