Beruflich Dokumente
Kultur Dokumente
Term 2, 2019
CVEN3031 CIVIL AND ENVIRONMENTAL ENGINEERING PRACTICE
Literature Review:
OD Matrix Estimation
and Light Rail Transit
This literature review will analyse aspects of origin-destination matrix estimation
methods and light rail transit as well as a bibliometric analysis on LRT.
Kalman Filtering
Kalman filtering is a popular algorithm that produces a solution using parameters that
can vary with time and observations that can include uncertainty (Harvey, 2009). It is an
incremental process where the results become updated and more accurate with additional
data. There are two main steps in the process:
- Estimating current variables using measurements and models
- Predict future variables based on previous measurements
Advantages
One of the most important advantages of Kalman filtering is the high accuracy of the
solution since it incorporates newer measurements in each iteration. As a result, it also is
able to create dynamic OD matrices where the matrix is subject to change with time,
which is highly useful in analysing real world scenarios and is a problem for simpler
models. Additionally, it is flexible in that measurements can be assigned different weights
if there were a set of bad measurements for example and thus would have a lesser effect
on the outcome.
Disadvantages
The processes involved in Kalman filtering requires high computational power due to the
intense linear algebra computations and the existence of sparse equations matrices. If this
method was applied to large scale problems, this would involve inverting and
manipulating large matrices which lengthen the processing time greatly. In addition, this
method has constant numerical complexity which means the computational power
required is the same regardless of the number of variables it has to solve meaning that
even small scale problems would require large computations. (Bierlaire & Crittin, 2004)
Another disadvantage is that if more data was obtained to improve the quality of the
matrix, equations have to be re-derived to accommodate for this change.
Least squares is an iterative process that minimises the residuals (differences) between
measurements and the proposed solution. (Harvey, 2009). There is a study that proposed
the use a generalised least squares estimator to derive OD matrices by minimising the
differences between the OD solution and a matrix obtained from a previous estimate or
model. (Cascetta, 194) Further research has led to improved least squares methods
including LSQR (Paige & Saunders, 1982) which was developed to be able to handle
sparse equations and matrices.
Advantages
An advantage of GLS is that many observations from a combination of sources are able
to be used in the process such as survey data and traffic count data. This means that the
program needs little change to accommodate different types of data. Additionally, like
Kalman filtering, each observation has an associated variance that can be changed
accordingly allowing the user to determine which measurements have more, or less,
weighting on the final result.
Another study proposed the use of LSQR, a version of least squares that is able to deal
with sparse equations and matrices efficiently, which are often present in the processes.
(Bierlaire & Crittin, 2004) The case study shows that the number of floating point
operations (flops), a measure of computational power, can be significantly lower than
Kalman filtering and still provide similar results. (Appendix A). Thus it is possible to
obtain results quickly and with less computations required.
Disadvantages
Despite the advantages, least squares does not produce the most accurate solutions
especially for LSQR. To add, for small values in an OD pair, the method can sometimes
produce negative numbers which is illogical and affects the overall solution. (Bell, 1991)
Although there are constraints proposed to remedy this, such as non-negativity or
inequality constraints, there has not been enough research into this.
Gravity Models
Gravity models, as the name implies, use a similar equation to Newton’s Law of
Gravitation to predict the number of trips in an OD pair. An example of a simple
formula used is shown below:
aPi Pj
Tij 2
d ij
Where Pi and Pj represent population at origin i/destination j, dij represents the distance
between origin and destination and α is a scale constant. (Ekowicaksono et al., 2016)
However with increasingly complex situations, the simple gravity model has been
modified to account for external influences and produce more accurate predictions. An
example is:
d
Tij bOi Di f (c) where f (c) cij
Where Oi, Di can represent any information related to origin i/destination j such as
population or employment. f(c) is called the deterrence function and represents the cost
of travel such as distance. b and d are calibration parameters. (Willumsen, 1981)
However, to calibrate the b and d parameters to suit the model, a regression method is
required to minimise the differences between modelled flows and observed flows. An
alternative is to use the Hyman method which is more robust and effective
(Ekowicaksono et al., 2016)
Advantages:
A significant advantage of the gravity model is that it is a very simple process compared
to other methods. For this reason, the process requires little computational power and
can produce results quickly. Furthermore, since the method is simple and requires little
resources, there are opportunities to use it in developing countries. (Willumsen, 1981)
The main reasons being that resources are scarce and conditions are often rapidly
changing so any long-term plans may become irrelevant prematurely.
Disadvantages:
The gravity model is a static method meaning that the OD matrix it creates does not
change with time. The results are still useful however this limits the analysis to a certain
time window. Additionally, due to its simplicity, it is not able to capture complex
scenarios and will produce inaccurate results. (Willumsen, 1981)
Neural Networks
Neural networks are complex systems that are modelled off biological neural systems.
The process involves layers consisting of neurons linked to neurons in other layers to
form a complex network. The first set of neurons become ‘activated’ upon a certain
condition which activates other neurons in the following layers depending on more
conditions. Eventually, the activated neurons in the last layer produce the final results.
(Marr, 2018) From this, the neural network is also able to perform machine learning
which allows it to perform tasks without explicit instructions but rather ‘learning’ it from
experience. By setting up the conditions appropriately, this method can be applied to OD
matrix estimation by using it to minimise the squared error between observed flow and
modelled flow. (Lorenzo & Matteo, 2013) The diagram below depicts what a neural
network could look like for estimating an OD matrix.
Advantages
The neural network is able to produce solutions in a shorter amount of time compared to
other methods. Moreover, it is a particularly robust method, allowing it to still produce
reasonable results even when there are bad/missing observations. (Lorenzo & Matteo,
2013) Furthermore, a time variable can be included to produce a dynamic OD matrix
which offers more for traffic analysis.
Disadvantages
A new neural network must be developed if the OD pairs change which change the OD
matrix structure. This could arise when the link between two centroids become no longer
accessible such as flooding events or vehicle accidents. A significant disadvantage is that
this method is a relatively new area of study so there is little research in the use of neural
networks for OD matrix estimation. Thus, it cannot be used in real world applications
yet however has shown to be of great potential to do so. (Remya & Mathew, 2013)
Conclusion
Although there are several OD estimation methods that already exist, research is still
ongoing to perfect the system. Each method has their own advantages and disadvantages
so its suitability is entirely dependent on the problem at hand and any additional
constraints. Only a few methodologies were reviewed but it was shown to have vastly
different techniques that provide good results however to solve real world problems, the
most appropriate one must be chosen based on circumstances.
Accessibility
In studies that observe the accessibility of light rail for people, the most popular method
is by analysing buffer zones around transit stops, which can be up to a one mile radius, as
well as travel distance between homes and the stop. (Hess & Almeida, 2007) Another
simpler method is to label houses as either inside or outside of the one-quarter or half a
mile radius of the transit station. This method would be more accommodated to large
area studies.
Another study proposed the use of park and ride facilities, offering car users to transfer
to LRT to continue onto their destination which is particularly advantageous if traffic
congestion occurs along the route or if parking facilities at the location are costly.
(Dickins, 1991) Theoretically, this expands accessibility to cover larger areas around the
station, assuming the users have access to cars, especially in areas where walking or other
forms of transport are impractical. This is also an opportunity for the user to save costs
as generally public transport is cheaper than paying for fuel. The success of park and ride
facilities are already evident in some North American cities however it has been shown
that it is difficult to predict demands and therefore the sizes of the facilities to
accommodate this demand - further research is required.
Overall, there has only been some research into the topic of accessibility itself but rather
how it leads on to affect property values which will be discussed later.
Additionally, a list of reasons were provided by another study that could explain a user’s
attraction to LRT (Scherer, 2010):
- Visual stimuli: vehicle designs, route visibility (tracks) and media coverage (especially
during the construction phase) enhances the perceptions and memories of LRT for
commuters. Additionally, since routes often run through attractive areas (such as
shopping districts), users psychologically link the transit with these areas.
- Reliability and comfort: LRT can be more reliable as it is less affected by traffic
especially since it has a dedicated ‘lane’ and right of way for most of the route. Moreover,
users experience more comfort on a LRT since the tracks are partly designed to provide
maximum comfort for passengers such as no sharp turning and flat grades.
- Large capacities: LRT can accommodate more passengers per vehicle compared to
buses, allowing larger flows of commuters throughout the network.
A study of LRT in Australia shows that despite little expansion of light rail networks in
the past decade, the number of passengers have increased 46% between 2001-2 and
2011-12 which was significantly higher than other modes of public transit. (Currie and
Burke, 2013) This could be due to the upgrading of old street car systems in place.
Further analysis shows that Melbourne is the actually leading contributor and contains
one of the largest light rail networks globally.
For countries that are relatively new to the LRT technologies, it is difficult to estimate
project costs without previous experiences. An example is shown by a study done in
Korea as there were no LRT operations at the time. As a result a Life Cycle Cost
estimator (LCC) was developed to estimate the construction and maintenance costs for
LRT which would greatly help in transport planning. (Kim et al., 2010) This same
estimator could be used in other countries however may not be as accurate as there are
differences in the laws regarding issues such as maintenance and construction.
Comparisons with bus rapid transit (BRT) and/or other public transport
modes
It appears that once implemented, LRT can be a better alternative than other transport
modes especially buses. A study in Portland showed that the impact of LRT led to
households in the outer portions of the network became less auto-oriented and were
more likely to use transit compared to the households in the bus network. (Dueker &
Bianco, 1999) Thus LRT appears to be more attractive for commuters than bus systems.
Furthermore, another study showed that the cost per passenger trip and per passenger
mile for LRT in four out of the five American cities studied were less than the respective
bus systems. (Black, 1993) To add on, LRT produces less regional emissions despite
improved bus emissions from better diesel technology. (Puchalsky, 2005) Thus LRT is a
possible alternative for high-volume bus systems.
However, from a city planning standpoint, there is significantly more planning, costs and
construction time involved when implementing LRT as it is a more permanent
infrastructure. Whereas for bus systems, routes and vehicle frequency can be changed
accordingly with minimal construction to accommodate changes in demand and thus
provide lower waiting times for customers. Buses also have lower operating costs
however operate at lower speeds compared to rail transit. (Tirachini, Hensher, &
Jara-Díaz, 2010)
Although LRT has shown to be able to improve the economy by providing new
accessibility to different areas, generally there are other elements that contribute
simultaneously to economic growth. (Knowles & Ferbrache, 2016) The effect can be
enhanced however by planning the transit system in tandem with land usages.
Additionally, a strong and growing economy that is already existing has shown to be one
of the important factors in determining whether the LRT has a positive impact on the
area. (Cervero, 1984)
In regards to health, the implementation of LRT has shown to be positive impact on the
users who walk to the station. A study in Portland has shown that there has been
evidence that there is a link between accessibility and increased physical activity as well as
decrease in chronic disease. (Frank et al., 2019) The methodology used to obtain this data
was to simply compare the health between two groups of LRT users, where one group is
close in proximity to the station and the other is not. The objective of this study was to
demonstrate how transport planners should also take health and well-being into account,
besides the economy and congestion relief, when planing for a mass transit system.
Safety
Safety is an important aspect for planners to consider especially if there is mixed traffic
(light rail and cars) as consequences have the potential to be severe. A study of the trams
in Melbourne discovered that, before modern stops, 82% of incidents near light rail stops
are actually between cars and pedestrians since the tracks are in the middle lane by
original design and commuters are required to cross a traffic lane/s to reach the tram, as
shown by pictures in Appendix D. (Currie & Reynolds, 2010) However, by altering the
design of the stops to promote safety, the study has shown it has been effective in
reducing the number of accidents with pedestrians. The designs of these modern stops
now include a dedicated raised platform, similar to train platforms, or the car lane where
pedestrians must cross is raised to act as a long speed hump (also shown in Appendix D).
However, due to the reduction of the lane width, this can become hazardous for vehicles
as well as cyclists. Ideally, the light rail tracks should be moved to the lane adjacent to the
curb to eliminate the issue of pedestrians crossing lanes however there are significant
costs in this proposal.
Conclusion
LRT has become a promising alternative to other modes of transport. However, where
there are positives there are also negatives and so LRT is not the only solution to mass
transit systems. But rather, it is a great accompaniment with existing public transport and
transport planners should consider it as a solution.
Evolution of Research
According to Figure 2 (below), which was obtained from Web of Science, there has been
an overall steady increase in research since 1995 with a huge peak at 2016. In these
records, the earliest research was published in 1976.
Interestingly, data from Google Trends shows a slow decrease in search interest
worldwide (Figure 3). However, in Australia, there is a clear positive trend starting from
2012-2013 which could be explained by governments’ interests in implementing LRT in
major cities such as Canberra and Sydney (more later).
Figure 3. Trend line for LRT search interest worldwide (top) and Australia (bottom) (Google Trends)
Geographic Significance
A significant portion of research was undertaken in the USA, followed by China, Canada,
Australian, England and Spain respectively as shown in the Figure 4 below. Light rail has
gained much momentum in the USA with the implementation of many projects across
the country and thus would propagate much research since data is readily abundant.
When analysing data from Google Trends, the regions with the most search interest are
Australia, USA, Hong Kong, New Zealand and Canada in descending order as shown in
Figure 5. More interestingly however is when analysing Australia only, ACT and NSW
produce the most searches. This can be accounted for due to recent and still ongoing
major light rail projects Canberra Metro (ACT) and Sydney Light Rail (NSW) which has
gained much traction.
Figure 5. Region distribution for LRT search interest worldwide (top) and Australia (bottom) (Google
Trends)
- Time: analysing the service times of light rails or the construction time to implement
LRT.
- Service: the quality of service that light rail provides compared to other modes of
transport.
- Model: some studies require the use of simulations and models to test their proposals
such as new efficient algorithms.
- City: often LRT is situated in cities as it is complimentary with the urban environment
and provides users with access to central business districts
- Bus: LRT is seen as an alternative to buses an therefore there have been some studies
comparing LRT with buses in terms of efficiency, cost, impact etc. This is important for
city planners to consider.
- Cost: determining project and maintenance costs of LRT are important for transport
planners and are one of the main considerations for a proposal
- Operation: determining a LRT success can be done by analysing its operations to observe
trends.
Figure 7 is a network visualisation that displays how key terms relate to each other. From
this, 3 main clusters can be observed:
- Red: key terms include service, development, model, environment, planning, location,
benefit, region and accessibility. This cluster appears to relate to analysing the issues and
benefits of LRT which can include the use of models to aid in the study.
- Blue: key terms include bus, cost, city, demand, quality, ridership and passenger. This
cluster observes the users of LRT and how they are impacted with regards to other
modes of transport and the quality of services for example.
Conclusion
In conclusion, the bibliometric analysis has shown that there has been extensive research
into LRT and studies are still ongoing. Most of the research is taking place in the USA as
LRT has been gaining popularity in the country. It has been proven to be a good form of
transport with good future potential for research and development.
Appendices
Appendix A: Estimated flops and residual comparison of the Irvine network (Bierlaire & Crittin,
2002)
Appendix B: Results from survey about satisfaction of the Hiawatha Light Rail (Cao, 2013)
Appendix C: Emmission levels of a) NOx, b) CO, c) PM10, and d) VOC (Messa, 2006)
Appendix D: Old light rail stops (top) vs modern light rail stops (middle and bottom) (Currie &
Reynolds, 2010)
References
OD Matrix Estimation
Babb, T. (2015). How a Kalman filter works, in pictures. [online] Bzarg. Available at:
https://www.bzarg.com/p/how-a-kalman-filter-works-in-pictures/ [Accessed 2 Jul.
2019].
Barceló Bugeda, J., Montero Mercadé, L., Bullejos, M., Serch, O. and Carmona Bautista,
C., 2012. A kalman filter approach for the estimation of time dependent od matrices
exploiting bluetooth traffic data collection. In TRB 91st Annual Meeting Compendium
of Papers DVD (pp. 1-16).
Bierlaire, M. and Crittin, F., 2004. An efficient algorithm for real-time estimation and
prediction of dynamic OD tables. Operations Research, 52(1), pp.116-127.
Cascetta, E., 1984. Estimation of trip matrices from traffic counts and survey data: a
generalized least squares estimator. Transportation Research Part B: Methodological,
18(4-5), pp.289-299.3
Cremer, M. and Keller, H., 1987. A new class of dynamic methods for the identification
of origin-destination flows. Transportation Research Part B: Methodological, 21(2),
pp.117-132
Harvey, B. (2009). Practical Least Squares and Statistics for Surveyors. 3rd ed. Sydney:
School of Civil and Environmental Engineering UNSW.
Hu, S.R., Madanat, S.M., Krogmeier, J.V. and Peeta, S., 2001. Estimation of dynamic
assignment matrices and OD demands using adaptive Kalman filtering. Journal of
Intelligent Transportation Systems, 6(3), pp.281-300.
Lorenzo, M. and Matteo, M., 2013. OD matrices network estimation from link counts by
neural networks. Journal of Transportation Systems Engineering and Information
Technology, 13(4), pp.84-92.
Marr, B. (2018). What Are Artificial Neural Networks. [online] Forbes. Available at:
https://www.forbes.com/sites/bernardmarr/2018/09/24/what-are-artificial-neural-net
works-a-simple-explanation-for-absolutely-anyone/#557bdc141245 [Accessed 4 Jul.
2019].
Paige, C.C. and Saunders, M.A., 1982. LSQR: An algorithm for sparse linear equations
and sparse least squares. ACM Transactions on Mathematical Software (TOMS), 8(1),
pp.43-71.
Remya, K.P. and Mathew, S., 2013. Od matrix estimation from link counts using artificial
neural network. CONTRIBUTORY PAPERS, p.287.
van der Zijpp, N., 1997. A comparison of methods for dynamic origin-destination matrix
estimation. IFAC Proceedings Volumes, 30(8), pp.1375-1380.
Light Rail
Black, A., 1993. The recent popularity of light rail transit in North America. Journal of
Planning Education and Research, 12(2), pp.150-159.
Cao, J., 2013. The association between light rail transit and satisfactions with travel and
life: evidence from Twin Cities. Transportation, 40(5), pp.921-933.
Cervero, R., 1984. Journal report: Light rail transit and urban development. Journal of
the American Planning Association, 50(2), pp.133-147.
Chen, H., Rufolo, A.M. and Dueker, K., 1997. Measuring the impact of light rail systems
on single family home values: A hedonic approach with GIS application.
Currie, G.V. and Burke, M., 2013, October. Light rail in Australia-Performance and
prospects. Australasian Transport Research Forum.
Currie, G. and Reynolds, J., 2010. Vehicle and pedestrian safety at light rail stops in
mixed traffic. Transportation Research Record, 2146(1), pp.26-34.
Dickins, I.S., 1991. Park and ride facilities on light rail transit systems. Transportation,
18(1), pp.23-36.
Dueker, K.J. and Bianco, M.J., 1999. Light-rail-transit impacts in Portland: The first ten
years. Transportation Research Record, 1685(1), pp.171-180.
Dure, D., 1999. Maximizing operating reliability in design of long single-track light rail
transit lines. Transportation research record, 1677(1), pp.73-78.
Frank, L.D., Kuntz, J.L., Chapman, J.E., Fox, E.H., Dickerson, J.F., Meenan, R.T.,
Saelens, B.E., Young, D.R., Boone-Heinonen, J. and Fortmann, S.P., 2019. The Health
and economic effects of light rail lines: design, methods, and protocol for a natural
experiment. BMC public health, 19(1), p.200.
Hess, D.B. and Almeida, T.M., 2007. Impact of proximity to light rail rapid transit on
station-area property values in Buffalo, New York. Urban studies, 44(5-6), pp.1041-1068.
Kim, G.T., Kim, K.T., Lee, D.H., Han, C.H., Kim, H.B. and Jun, J.T., 2010.
Development of a life cycle cost estimate system for structures of light rail transit
infrastructure. Automation in Construction, 19(3), pp.308-325.
Knowles, R.D. and Ferbrache, F., 2016. Evaluation of wider economic impacts of light
rail investment on cities. Journal of Transport Geography, 54, pp.430-439.
Messa, C.A., 2006. Comparison of emissions from light rail transit, electric commuter rail,
and diesel multiple units. Transportation research record, 1955(1), pp.26-33.
Puchalsky, C.M., 2005. Comparison of emissions from light rail transit and bus rapid
transit. Transportation research record, 1927(1), pp.31-37.
Scherer, M., 2010. Is light rail more attractive to users than bus transit?: Arguments based
on cognition and rational choice. Transportation research record, 2144(1), pp.11-19.
Tirachini, A., Hensher, D.A. and Jara-Díaz, S.R., 2010. Comparing operator and users
costs of light rail, heavy rail and bus rapid transit over a radial public transport network.
Research in transportation economics, 29(1), pp.231-242.
Topalovic, P., Carter, J., Topalovic, M. and Krantzberg, G., 2012. Light rail transit in
Hamilton: Health, environmental and economic impact analysis. Social Indicators
Research, 108(2), pp.329-350.