Vous êtes sur la page 1sur 6

CE 6870

TRANSPORTATION SYSTEMS ANALYSIS

Term Paper report


On

Estimation of Short-Term Bus Travel Time by Using Low-


Resolution Automated Vehicle Location Data

By

Yashar Zeinali Farid, Eleni Christofa, and Laurel Paget-Seekins

Submitted by,

B.Uday sagar,

CE18M070,

MTech –Transportation,

Civil Engineering Department

Indian Institute of Technology,Madras.

1
Estimation of Short-Term Bus Travel Time by Using Low-
Resolution Automated Vehicle Location Data

Background

Automatic Vehicle location data systems are those systems that provide real-time information
in the form of low resolution of data. The low resolution of data (data obtained every 60
seconds in this study) causes problem in predicting the trajectories of the routes which bus
has taken, even though, the cost of using such devices is very less compared to other
advanced technologies that give high resolution data, typically location of the bus in every
second. Several well-established techniques for data collection to estimate the travel time are
available at local and network level. At local level, inductive loops and infrared rays can be
used to detect vehicles near intersections, while at a network level, systems that give the
trajectories of the buses are essential. As explained earlier, AVL systems can be used for this
purpose. They have to be fit into buses and with the data provided by them, it is possible to
determine the bus routes with reasonable accuracy. However, if the systems are used in urban
routes where the links are very small and the buses are expected to traverse across a link in
less than 60 seconds, the problem of estimating the trajectories accurately will be difficult.
Estimation of bus travel time accurately using AVL data is very essential as many transit
operators use these devices. The study aims to obtain trajectories of buses accurately by using
the model-driven approaches for predicting travel time and they also compare the results of
model-driven approach with data-driven approach.

Need for the Study

Estimation of the bus travel times is an essential exercise for the transit agencies to monitor
the movement of buses for managing their fleets better and to provide their passengers with
real-time information on transit arrivals. Other advantages include efficient signal-timing
optimisation with transit signal priority.

Many transit operators use Automatic Vehicle Location (AVL) data for this purpose.
However, AVL data is generally obtained at very low frequency (Polling intervals that are
considered to offer high resolution are 1s), and often because of communication constraints
getting bus trajectory between two records exactly is very difficult. Thus, the trajectories

2
must be estimated accurately between the records and this also helps in estimation of bus
travel time accurately.

Thus, the objective of this study was to address inaccuracies in bus trajectory inference at
urban signalized arterials when only low-resolution time-based AVL data are available by
decomposing bus travel time to running travel time, intersection delay, and bus stop dwell
time.

Methodology

Several models used low frequency AVL data along with knowledge of network
specifications such as locations of bus stops and intersections. A segment of Washington
street in Boston, Massachusetts is selected, and a model-driven model (based on traffic flow
theory, and then unobserved variables are estimated by various techniques including Kalman
filtering, ML, and OLS) was fit. Bus travel time at signalized urban arterials was first
decomposed into three components: running travel time, bus stop dwell time, and intersection
delay. A linear regression model (Equation 1 shows the specification and Table 1 shows the
model) that estimates bus travel time at a certain location was estimated using real-world
AVL data. Then, various data-driven models (Use historical data, real-time data, or based on
density, flow, time of day, dwell times at bus-stops and weather conditions) were developed
by using the same AVL data to estimate bus travel time, and their predictions were compared
with those from the model-driven linear regression. The SVR model outperforms every other
model in generalized error measures like MAPE, RMSE etc as in Table 2.

𝐤,𝐛
𝐭̂ 𝐛𝐢 − 𝐭 𝐛𝐢−𝟏 = 𝛃𝟎 + 𝛃𝟏 × 𝐝𝐛𝐢−𝟏,𝐢 + ∑ 𝛃𝐤𝟐 × 𝐲𝐢−𝟏,𝟏 + ∑ 𝛃𝐧𝟑 × 𝐮𝐧,𝐛
𝐢−𝟏,𝐢
𝐤∈𝐈𝐊 𝐧∈𝐈𝐍

∀𝐛 ∈ 𝐈𝐁 ; 𝐢 ∈ 𝐈𝐛 ; 𝐢 ≠ 𝟏

Where, 𝒚𝒌,𝒃 𝒏,𝒃


𝒊−𝟏,𝒊 and 𝒖𝒊−𝟏,𝒊 are dummy variables indication the absence or presence of bus-stops

and intersections between records i-1 and i.

3
Results

Table 1 Linear Regression Model Parameter Estimates

Characteristic by β1 β21 Bus Stop 1 β22 Bus Stop 2 β31 Average


Model Average Average Dwell Average Dwell Intersection
Pace time (s) time (s) Delay
(s/mi)
Off-peak Period
Coefficient 266.93 18.17 16.21 40.05
t-statistic 14.96 8.91 8.96 18.97
p-value 0.00 0.00 0.00 0.00
Peak Period
Coefficient 305.15 19.43 15.56 37.65
t-statistic 11.63 6.64 5.81 12.21
p-value 0.00 0.00 0.00 0.00

Table 2 Generalised Errors of Models

Model Off-peak Peak


RMSE MAE RMSE MAE
Linear Regression 14.20 10.84 15.40 8.52
SVR 13.85 10.12 9.63 7.23
Decision Tree Regression 15.87 27.00 16.12 9.56
Lasso 14.20 10.14 12.55 7.53
Ridge 14.34 10.85 13.03 7.64
Bayesian Ridge 14.10 11.10 12.77 8.47

4
Strengths of the study

The study is very essential as estimation of the trajectories of the buses accurately when AVL
systems are used helps reduce the cost of the transit agencies to monitor buses and in
providing real time information about the bus arrivals and departures for the passengers. The
local traffic authorities also get benefitted from the real-time information about the buses for
the pre-emption of green signal for buses.

The study used simple linear regression model, which can be understood very easily. Also,
the division of the travel time into running travel time, dwell time at bus stops that may be
located between points and delays at signalized intersections. This helps in understanding the
effect of the individual components in the total travel time and thus helps in improving the
particular component for the reduction in overall travel time. For example, if the delay at
signalised intersections has more effect on the total delay and thus increased travel time, then
signalised intersections can alone be improved for the overall reduction in travel time.

Deficiencies of the study

The study however has certain deficiencies like it divides the travel time into only three
components. The running travel time is a function of several factors like presence of cross-
streets in the mid-block region, changes in road width etc. To consider this phenomenon into
account, the researchers can consider homogenous sections into account and divide the total
travel time into more components.

Also, the study is restricted to a stretch of road in Washington area. The spatial and temporal
transferability of the model is a question as in some other region, the number of signals may
be more along the stretch and this may lead to wrong estimation of the effect of delay at
signalised intersections on the total travel time. The temporal transferability is also a question
as signals will be of different cycle lengths across the day and thus the effect may be different
at different times.

If traffic signals are not present then the estimation will be completely different from what is
seen in this study.

5
Contribution of the Paper

The findings of this study are expected to improve traveller information systems and make
TSP strategies efficient improving the transit quality of service. It also reduces the cost
incurred by transit operators in finding the accurate locations of buses and in providing
accurate information to passengers. The findings make the entire transportation system
efficient by providing necessary transit information to the transit operators.

Future Scope

The travel time can be estimated using other advanced model-driven approaches like non-
linear least squares, generalised least squares etc. These advanced models are expected to
improve the accuracy in estimating travel time as the relation may not be linear.

A more generalised model that estimates travel time in all possible settings should be made.
This can be done by decomposing the study sections into further smaller homogenous
sections and this may improve the applications of the study.

Vous aimerez peut-être aussi