Article citation information:
Munshi, A.K.,
Patnaik, A.K. Modelling roundabout entry capacity for mixed traffic flow using
ANN: a case study in India. Scientific Journal of Silesian University
of Technology. Series Transport. 2024, 123, 209-226. ISSN: 0209-3324. DOI: https://doi.org/10.20858/sjsutst.2024.123.10.
Aarohi Kumar MUNSHI[1],
Ashish Kumar PATNAIK[2]
MODELLING ROUNDABOUT ENTRY CAPACITY FOR MIXED TRAFFIC FLOW USING ANN: A CASE
STUDY IN INDIA
Summary. Roundabouts,
as an unsignalized intersection, have an effective preventative measure
designed to control straight-line crashes. Efficient traffic flow in cities
depends upon appropriate capacity estimation of roundabouts. This study
attempts to develop models for roundabout entry capacity by applying Artificial
Neural Network (ANN) analysis for mixed traffic flow conditions. Data was
gathered from 27 roundabouts spread across India. The influence area for gap
acceptance (INAGA) concept was used as a graphical method to identify critical
gap (Tc) of entry flow at
roundabouts. This study indicated that the Bayesian Regularisation Neural
Network (BRNN) based model has the best R2 and RMSE of 0.97 and
167.8. The connection weight approach and Garson algorithm evaluate the
significance of each explanatory variable and identify follow-up time (Tf) as a critical parameter
with values of 11.10 and 21.15%, respectively.
Keywords: INAGA,
ANN entry capacity, Garson algorithm
1. INTRODUCTION
Roundabout
is an efficient traffic control measure in the form of safety and operational
aspects. Circular intersections substantially mitigate head-on collisions,
notably right-angle accidents, as well as facilitate the easy movement of more
traffic with reduced queuing time. In comparison to signalized junctions, it
has fewer conflict points. At signalized intersections, 32 potential sources of
conflict are identified, but at roundabouts, the number falls to 8. As a
result, roundabouts are regarded to be vital infrastructure in any country's
urban traffic system. The roundabout capacity is a key statistic for evaluating
operational performance, delay, and queue length.
The
upholding of driver conduct, specifically lane discipline, is inadequately
managed in conditions of heterogeneous traffic. In addition, the dimensions,
transmission capacities of vehicles, and geometrical features of intersections
in developing countries are very different from those of vehicles in affluent
nations. Drivers get aggressive while attempting to manoeuvre around one
another in congested areas. In such conditions, vehicle speeds vary as a
result of the inherent differences between vehicles. At roundabouts,
significant contributing vehicles, such as small-sized cars (SC) and
two-wheelers (2W), consistently seek to merge into
the mainstream flow of traffic. This results in an extremely
chaotic traffic situation due to the inconsistent spacing between vehicles. HCM
(2010) presents a comprehensive methodology for estimating the entrance
capacity of roundabouts, however, it fails to account for people's driving
behaviour under local conditions. Hence, emphasis has been given to develop the
capacity models for roundabouts under constrained conditions.
Essential
methodologies, such as empirical models, gap acceptance models, and microscopic
simulation modelling, serve as a basis for existing roundabout capacity models.
Empirical models are often solved by applying regression analysis, with
capacity serving as the dependent variable and other factors, such as
prevailing flow and site geometry, contributing as explanatory variables [1].
It has been identified that the entry capacity varies negatively exponentially
about the opposing flow and that as the opposing flow increases, the entry
capacity decreases and vice versa [2], [3]. The capacity model on the concept
of gap acceptance theory reflects the decision of drivers on variables like
critical gap and follow-up time. For drivers in the minor stream, the minimum
acceptable duration before merging into the major stream is termed as the
‘critical gap’ while the ‘follow-up time’ refers to the
time difference between vehicles in a queue during congested traffic conditions
[4]. In contrast, microscopic simulation models are influenced by the
interactions and motions of vehicles within a simulated network. Three primary
terms can be used to describe the actions of vehicles on the route:
car-following, lane-changing, and the gap acceptance concept, which includes
critical gap and follow-up time. Various microscopic simulation models have
been developed for parametric findings, with the flows and turns being
controlled [5], [6]. Artificial neural networks (ANNs) are frequently used by
professionals in academia, as they can model complicated and nonlinear data
sets accurately and with the least amount of error [7]. The capacity prediction
of roundabouts through ANN outperforms the gap acceptance and empirical models,
further it encourages employing machine learning techniques for a proactive
operational plan for roundabouts [8], [9]. Accordingly, an ANN-based entrance
capacity model for Indian urban cities has been developed.
Data
was obtained from 27 roundabouts in India during peak traffic hours. The
graphical method of influence area, INAGA, is used to predict driver response
factors such as critical gap (
The paper is
structured into four sections, with section 1 providing an overview of the
context of the study and a review of the relevant literature. The
subsequent section discusses study sites and data-collecting procedures. The
methodology and analysis of the ANN capacity models are
comprehensively described in Section 3. The final section of the paper provides
the conclusion of this study. It also addresses the limitations of the study
and suggests avenues for future research.
2.
SELECTION OF STUDY SITES AND DATA COLLECTION PROCEDURE
A total of 27
roundabouts were selected for data collection to develop capacity models. The
twelve cities depicted in Fig. 1 in the four corners of India all consist of
roundabouts. The roundabouts were selected following the specified criteria.
Roundabouts, which have 4 approach legs, tend to be placed in commercial,
industrial, and residential areas. The roundabouts tend to be at grade
intersections and are unsignalized. Further, the influence of cyclists and
pedestrians is modest at these roundabouts. The measured geometric
specifications of roundabouts are presented in Appendix 1.
Fig. 1. Location of Cities on India Map
The collection
of traffic flow data is from high-rise buildings in close proximity to the
roundabouts. This approach was taken to ensure the acquisition of comprehensive
and relevant data sets. The videography method utilized in this study is
straightforward and economical. It is also convenient to use a post-processing
method for data extraction. Videos capturing the morning and evening traffic
flow peaks under clear weather conditions were utilized to account for the
significant contribution of both the critical gap (
Tab.
1
Analysis of contributing variables for this study
|
Variables |
Units |
Minimum |
Maximum |
Mean |
Standard deviation |
|
Observed entry capacity ( |
PCU/h |
251 |
3346 |
1712.5 |
772.75 |
|
Circulating flow ( |
PCU/h |
220 |
3778 |
1051.28 |
711.42 |
|
Weaving length ( |
m |
29.76 |
59.52 |
43.39 |
6.09 |
|
Weaving width ( |
m |
9.1 |
19.04 |
16.80 |
5.34 |
|
Entry width ( |
m |
6 |
21.22 |
14.17 |
3.87 |
|
Diameter of central
island ( |
m |
12.66 |
62.32 |
42.62 |
11.77 |
|
Critical gap ( |
seconds |
0.68 |
2.66 |
1.73 |
0.58 |
|
Follow-up time ( |
seconds |
0.88 |
2.34 |
1.78 |
0.52 |
3. METHODS AND ANALYSIS
This section
discusses extensively the Gap acceptance variable and the existing models'
viability. Furthermore, the ANN model is being developed for the prediction of
roundabout entry capacity. Additionally, the entry capacity model's suitability
and the relative significance of the input variables are specified.
3.1.
Gap Acceptance Variables
For capacity
models of roundabouts, significant factors that are considered to describe
driver conduct under real-world traffic conditions include critical gap and
follow-up time. An inadequate approach for obtaining critical gap and follow-up
time values results in partial capacity estimation of a roundabout. In order to
determine the critical gap, both the newly accepted equilibrium probability
method and the widely utilized Raff method considered homogeneous
traffic flow and consistent driving conduct. Nevertheless, with a zero
rejection of gap data, these approaches fail to produce reliable results.
Therefore, it is presumed that this is a significant issue in mixed traffic
conditions. In mixed traffic situations, zero gaps are typically rejected for
congested and normal traffic flow. Subsequently, to surmount these
shortcomings and variations, the INAGA method, which was recently devised, is
implemented to produce dependable outcomes [10]. The INAGA technique can
calculate the critical gap without having a specific distribution function as
an assumption. As a result, the INAGA method is a graphical method of
assumption; upon observing the mainstream where the merging of
entry and circulating flows is most, a trapezoidal-shaped influencing area may
be meticulously assumed.
3.2.
Existing Model Evaluation
The Girabase
formula (France), Brilon-Wu formula (Germany), and HCM 2010 (USA) models are
used for a new set of roundabouts to evaluate their ability to anticipate
traffic flow. In a study, the feasibility of using explanatory variables to
establish a relationship between capacity, other geometric functions, and gap
acceptance has been analysed [11]. As illustrated in Fig. 2, the aforementioned
techniques generate a relation between the data obtained from field
observations and the predicted capacity.
(a)
(b)
(c)
Fig. 2. Prediction of capacity by using
international capacity models
Figures
2(b) and (c) show that, in comparison to capacities observed under field
conditions, the predicted entry capacities of the Brilon-Wu formula (Germany)
and the HCM 2010 approaches are, respectively, higher, and lower. This shift in
capacity estimates can be attributed to a couple of factors. The first cause is the fact that metropolitan Indian
streets see a wide range of vehicle types. The second issue is likely the
vastly different driving conduct of Indian drivers compared to those of drivers
in Western countries. In mixed traffic flow, vehicles with
varied manoeuvrability coexist. The prevalence of two-wheelers over
cars of all kinds predominates in Indian traffic. A breach of the principle of
priority happens whenever two-wheelers (2W) in the approaching stream
emphatically establish a gap in the route of the primary traffic flow.
Homogeneous traffic, on the other hand, involves vehicles moving in a
streamlined pattern whilst consistently maintaining some distance. When applied
to data sets containing mixed traffic flows, models that were fitted under
homogeneous traffic conditions overestimate entry capacity under high
circulating flows, resulting in inaccurate model fitting. Measurements of
driver behaviour metrics, such as critical gap and follow-up time, under actual
conditions, might help reduce discrepancies in capacity projections.
Furthermore, the significance of geometric variables in capacity prediction
analysis is acknowledged. By integrating geometric and gap acceptance variables
as explanatory variables in the context of mixed traffic flow conditions, gap
acceptance capacity models are subsequently devised. However, the capacity
values of the proposed Bayesian Regularisation Neural Network (BRNN) model and
the Girabase formula in Fig. 2(a) are comparable.
3.3.
Artificial Neural Network (ANN) Modelling
Modern data
analysis techniques, such as ANN modelling, offer an alternative to the
traditional statistical regression method. A linear model aims to establish a
linear relationship between independent and dependent variables. However, in
ANN modelling, the coefficient, and intercept variables are associated with
neural network weights and biases. The hidden layers in the neural network are
accomplished to evaluate the critical relationships, such as the nonlinear
relationship between dependent and independent variables. For estimating
roundabout capacity, the authors discovered that ANN modelling outperformed
regression models [1]. In regression models, there is a probability of the
occurrence of constraints between explanatory variables and capacity. However,
ANN modelling can establish an appropriate relationship between explanatory
variables and capacity even if there is a constraint relationship between
explanatory variables and capacity.
The current
study employs two distinct training algorithms, namely the Bayesian Regularisation Neural Network (BRNN) and
Levenberg-Marquardt Neural Network (LMNN)
to develop the ANN models. As illustrated in
Fig. 3, feed-forward perceptron with back propagation training algorithm (FFBP)
type of ANN with hyperbolic tangent activation functions is developed using explanatory
variables like weaving length (
Fig. 3. Neural network diagram with 7 explanatory variables
A total of 110
data points were used to forecast the ANN capacity models. The entire dataset
was split in half, with 70% being used for the training phase and 30% for the
testing phase. Several iterations are performed to determine the best ANN capacity
model, as detailed in Tab. 2. the best ten ANN models are selected through
root-mean-square error (RMSE) and overall R2 values. Finally, the
most suitable ANN model is selected from ten ANN models. In this study, three Levenberg-Marquardt Neural Network
(LMNN) and seven Bayesian Regularisation Neural Network (BRNN) based capacity
models are developed. The transfer function introduces non-linearity to the
neural network. It transforms the weighted sum of inputs and biases into the
output of a neuron. In the present model, the hyperbolic tangent sigmoid
transfer function (Tansig) squashes input values to a range between -1 and 1.
It is a smooth, differentiable function, crucial for backpropagation during
training. The training function Bayesian regularisation back propagation (Train
BR) is responsible for adjusting the weights and biases of the network during
the training phase. Its goal is to minimize the difference between the
predicted outputs and the observed values. The optimisation algorithms use the
gradient of the loss function concerning the parameters to update the weights
and biases iteratively. The learning function involves setting hyperparameters,
including the learning rate and possibly other parameters that control the
training process. The learning rate determines the step size during weight
updates and influences the convergence and stability of the training, and hence
an adaptive learning function gradient descent with momentum (Learn GDM) is
employed for the study. The performance function quantifies the difference
between the predicted values and the observed values. Consequently, mean square
error with regularisation (MSEREG) is a performance function that combines the
Mean Square Error (MSE) with a regularisation term to prevent overfitting and
control the complexity of a model during training and predicts the capacity of
a developed model by using ANN. In addition to these, the number of iterations
(epochs) varied from 500 to 1000 to get the best result. Based on the RMSE
results and overall R2, the most suitable model is selected. It is
observed from Tab. 2 that BRNN based model having four numbers of hidden
neurons is the best fitted as the RMSE and overall R2 values of the
model were found to be 167.89 and 0.97 respectively.
Tab.
2
Network details with
several iterations (Trial 1-10)
|
Trial No. |
1 |
2 |
3 |
4 |
5 |
|
No. of hidden layer |
1 |
1 |
1 |
1 |
1 |
|
No. of hidden
neurons |
3 |
4 |
6 |
7 |
8 |
|
Transfer function |
Tansig |
Tansig |
Tansig |
Tansig |
Tansig |
|
Training function |
Train LM |
Train BR |
Train BR |
Train BR |
Train BR |
|
Adaptive learning
function |
Learn GDM |
Learn GDM |
Learn GDM |
Learn GD |
Learn GDM |
|
Performance
function |
MSE |
MSEREG |
MSEREG |
MSEREG |
MSEREG |
|
No. of epochs |
500 |
1000 |
1000 |
700 |
1000 |
|
RMSE |
267.28 |
167.8 |
212.21 |
216.63 |
230.30 |
|
Overall R2 |
0.92 |
0.97 |
0.96 |
0.96 |
0.95 |
|
Trial No. |
6 |
7 |
8 |
9 |
10 |
|
No. of hidden layer |
1 |
1 |
1 |
1 |
1 |
|
No. of hidden
neurons |
3 |
4 |
7 |
6 |
8 |
|
Transfer function |
Logsig |
Logsig |
Logsig |
Logsig |
Logsig |
|
Training function |
Train BR |
Train BR |
Train BR |
Train LM |
Train LM |
|
Adaptive learning
function |
Learn GDM |
Learn GD |
Learn GDM |
Learn GDM |
Learn GD |
|
Performance
function |
MSEREG |
MSE |
MSEREG |
MSE |
MSE |
|
No. of epochs |
500 |
1000 |
1000 |
1000 |
700 |
|
RMSE |
222.14 |
190.3 |
214.00 |
218.2 |
240.9 |
|
Overall R2 |
0.95 |
0.96 |
0.96 |
0.95 |
0.95 |
Note:
Tansig:
Hyperbolic tangent sigmoid transfer function
Logsig: Log-sigmoid transfer function
Train LM: Levenberg-Marquardt backpropagation training
function
Train BR: Bayesian regularization backpropagation
training function
Learn GD: Gradient descent adaptive learning function
Learn GDM: Gradient descent with momentum
MSE: Mean square error
MSEREG: Mean square error with regularization
An ANN
equation was formulated in the present study by utilizing 7
explanatory variables and observing through a trained network. By using the
weights and biases observed in this analysis as provided in Tab. 3, the
following mathematical expressions were developed for the capacity prediction
of roundabouts. For this, the procedure follows three steps. In the first step,
Tab. 3
Connecting weights and
biases with normalized entry capacity prediction model
|
No. of hidden neurons |
Weights |
Biases |
||||||||
|
I1 ( |
I2 ( |
I3 ( |
I4 ( |
I5 ( |
I6 ( |
I7 ( |
O ( |
I |
O |
|
|
1 |
-0.068 |
-0.539 |
0.298 |
0.261 |
-0.563 |
-0.537 |
-0.348 |
1.040 |
0.142 |
0.389 |
|
2 |
0.221 |
-0.329 |
0.003 |
0.545 |
-0.560 |
0.528 |
-0.343 |
0.697 |
-0.602 |
- |
|
3 |
0.388 |
0.282 |
0.029 |
-0.262 |
-0.309 |
-0.650 |
0.250 |
0.873 |
-1.450 |
- |
|
4 |
-0.061 |
-1.184 |
0.047 |
-0.285 |
-0.752 |
-0.497 |
-0.913 |
-0.772 |
-0.021 |
- |
Tab. 4
Connecting weights of four hidden neurons
|
No. of hidden neurons |
Weights |
|||||||
|
I1 ( |
I2 ( |
I3 ( |
I4 ( |
I5 ( |
I6 ( |
I7 ( |
O ( |
|
|
1 |
-0.0682 |
-0.5399 |
0.2984 |
0.2619 |
-0.5635 |
-0.5374 |
-0.3486 |
1.0402 |
|
2 |
0.2217 |
-0.329 |
0.0031 |
0.5453 |
-0.5607 |
0.5281 |
-0.3439 |
0.6974 |
|
3 |
0.3888 |
0.2827 |
0.0295 |
-0.2624 |
-0.3095 |
-0.6504 |
0.2506 |
0.8736 |
|
4 |
-0.0611 |
-1.184 |
0.0474 |
-0.2852 |
-0.7529 |
-0.4978 |
-0.9136 |
-0.7726 |
The expression for evaluating B term is
written as follows:
Then the final term
The
aforementioned equation (4) yields a capacity value between -1 and 1, which
will be denormalized according to equation (5).
Where,
3.4. Suitability of entry capacity model
To ensure the
development of a suitable entry capacity model, a number of statistical tests
are performed to evaluate the reliability of the ANN model as detailed
in Tab. 5. The statistical tests include best-fit calculations,
error-calculating variables, mathematical calculations, cumulative probability
values, and predictions with an accuracy level of less than 20%. The coefficient of determination (R2),
which executes from 0 to 1, represents the degree of goodness of fit, and
higher values indicate a better fit. Nash-sutcliffe model efficiency coefficient (E) is generally used to
assess the prediction ability of the developed model. The value of ‘E’ can range
from -∞ to 1. If the value of ‘E’ is close to 1, then it is
known to be more accurate in the developed model. The absolute value is
measured by the modulus value of the difference between observed and predicted
values. The average absolute error (AAE) is measured as the average of the
absolute difference between observed and predicted values, whereas the maximum
of the measured absolute errors is known as maximum absolute error (MAE). The Root Mean Square Error (RMSE) is intended to indicate a model's
accuracy, and a smaller value is preferable. The ranges of
AAE and RMSE are in between 0 to ∞. The detailed formula for evaluating
the R2, E, AAE, MAE and RMSE are given as equations 6(a) to 6(e). The
Tab. 5
Statistical test of developed model
|
Model |
Data Splitting |
Co-relation
Analysis |
Error Calculations |
Mathematical Calculations of CP/CO |
Cumulative Probability of CP/CO |
± 20%
Prediction Accuracy (%) |
||||||
|
R2 |
E |
AAE |
MAE |
RMSE |
µ |
σ |
Ratio* at |
Log- normal |
Histogram |
|||
|
P50 |
P90 |
|||||||||||
|
BRNN Model |
Training |
0.93 |
0.93 |
149.2 |
455.1 |
184.8 |
1.00 |
0.11 |
0.99 |
1.13 |
92.32% |
100% |
|
Testing |
0.95 |
0.94 |
167.9 |
418.7 |
197.9 |
1.05 |
0.22 |
0.95 |
1.22 |
91.01% |
94% |
|
Note:
R2 = Coefficient of determination, E =
Nash-Sutcliffe coefficient, AAE = Average Absolute
Error, MAE = Maximum Absolute Error, RMSE = Root Mean Square Error, CP = Predicted Capacity, CO =
Observed Capacity
3.5. Relative Importance of Input
Variables
To assess the
relative importance of input variables in ANN modelling, two methods such as
Connection weight approach and Garson’s algorithm are applied in this
study. In connection weight approach, the product of each input and output
weights (
Tab. 6
Products of each input and output weights (
|
No. of hidden neurons |
I1 ( |
I2 ( |
I3 ( |
I4 ( |
I5 ( |
I6 ( |
I7 ( |
|
1 |
-0.0709 |
-0.5616 |
0.3103 |
0.2724 |
-0.5861 |
-0.559 |
-0.3626 |
|
2 |
0.1546 |
-0.2294 |
0.0021 |
0.3802 |
-0.391 |
0.3682 |
-0.2398 |
|
3 |
0.3396 |
0.2469 |
0.0257 |
-0.2292 |
-0.2703 |
-0.5681 |
0.2189 |
|
4 |
0.0472 |
0.9147 |
-0.0366 |
0.2203 |
0.5816 |
0.3846 |
0.7058 |
Tab.
7
Ratio of (
|
No. of hidden neurons |
I1 ( |
I2 ( |
I3 ( |
I4 ( |
I5 ( |
I6 ( |
I7 ( |
|
1 |
0.0455 |
0.3605 |
-0.1992 |
-0.1749 |
0.3763 |
0.3589 |
0.2328 |
|
2 |
3.4432 |
-5.1091 |
0.0467 |
8.4677 |
-8.7082 |
8.2004 |
-5.3407 |
|
3 |
-1.4359 |
-1.0439 |
-0.1086 |
0.9691 |
1.1429 |
2.4021 |
-0.9255 |
|
4 |
0.0167 |
0.3246 |
-0.0129 |
0.0781 |
0.2064 |
0.1364 |
0.2504 |
|
Sum |
|
|
|
|
|
|
|
The analysis
suggests that the BRNN model is best for the present study. Additionally, this
study applied the Connection weight technique and Garson's algorithm for
assessing the relevance of explanatory variables in ANN modelling [12].
The input variables' contributions are detailed in Tab. 8. The Connected
Weights technique, as given in equation 7(a), allows greater influence on input
variables with larger absolute weights on the output calculation of the hidden
layer and the overall performance of the network. Garson's algorithm employs
the same procedure to assess the contribution of input variables, and the
ranking of each input variable is determined by equation 7(b).
Where S1,
S2, ……S7 is the sum of product in
each hidden neuron.
Tab. 8
presents the ranking of input variables based on the absolute value of
Connection weight. Follow-up time (
Tab. 8
Contribution of input variables in BRNN model
|
Input variables |
Connection weight approach |
Garson algorithm |
||
|
Sum (Absolute Value) |
Rank |
Relative importance (%) |
Rank |
|
|
I1 ( |
2.07 |
6 |
7.72 |
6 |
|
I2 ( |
5.47 |
5 |
19.56 |
2 |
|
I3 ( |
0.27 |
7 |
3.53 |
7 |
|
I4 ( |
9.34 |
2 |
12.80 |
5 |
|
I5 ( |
6.98 |
3 |
19.50 |
3 |
|
I6 ( |
11.10 |
1 |
21.15 |
1 |
|
I7 ( |
5.78 |
4 |
15.71 |
4 |
ANN-based model has been
receiving wide appreciation over regression-based models as it is capable of
establishing nonlinear relationship between dependent and independent
variables. In this study, ten ANN-based models were developed for roundabout
entry capacity prediction purpose. It was observed that the BRNN based model
has the highest R2 value of 0.97 and lowest RMSE value of 167.89
among all ten models. Therefore, this model was selected for the capacity
prediction in this study. The comparison of various existing capacity models
with the ANN model is depicted in Fig. 4.
Moreover, to appraise the BRNN
model, several statistical tests were performed under a given data set.
Sensitivity analysis is carried out using Connection weight approach and Garson
algorithm to observe the influence of input variables in the proposed BRNN
model according to Garson algorithm, gap acceptance variables including
follow-up time (
Planners
and designers now have a practical tool in the form of the proposed models
(ANN) for predicting capacity under traffic situations analogous to those in
other developing countries; however, further research is needed to
determine the effect of pedestrian crossings on this estimation.
Fig. 4. Comparison of various capacity models
References
1.
Yap Yok Hoe, Helen M
Gibson, Ben J Waterson. 2013. “An International Review of Roundabout
Capacity Modelling”. Transport
Reviews 33(5): 593-616. DOI: https://doi.org/10.1080/01441647.2013.830160.
2.
Hamim Omar Faruqe,
Md. Saddam Hossain, Md. Hadiuzzaman. 2022. “Developing Empirical Model
with Graphical Tool to Estimate and Predict Capacity of Rural Highway
Roundabouts”. International Journal of Transportation
Science and Technology 11(4): 726-737. DOI: https://doi.org/10.1016/j.ijtst.2021.10.002.
3.
Ahmad Abdullah, Rajat Rastogi. 2016. “Regression
model for entry capacity of a roundabout under mixed traffic condition –
an Indian case study”. Transportation
Letters 9(5): 243-257. DOI: https://doi.org/10.1080/19427867.2016.1203603.
4.
Qu Xiaobo, Jin Zhang, Shuaian Wang, Zhiyuan Liu. 2014.
“Modelling follow up time at a single-lane roundabout”. Journal of Traffic and Transportation
Engineering 1(2): 97-102. DOI:
https://doi.org/10.1016/S2095-7564(15)30093-3.
5.
Fortuijn Lambertus G.H. 2009. “Turbo
Roundabouts: Estimation of Capacity”. Transportation
Research Record 2130(1): 83-92. DOI: https://doi.org/10.3141/2130-11.
6.
Valdez Marilyn, Ruey Long Cheu, Carlos Duran. 2011
“Operations of Modern Roundabout with Unbalanced Approach Volumes”.
Transportation Research Record
2265(1): 234-243. DOI: https://doi.org/10.3141/2265-26.
7.
Karlaftis M.G., E.I. Vlahogianni. 2011.
“Statistical methods versus neural networks in transportation research:
Differences, similarities and some insights”. Transportation Research Part C: Emerging Technologies 19(3): 387-399.
DOI: https://doi.org/10.1016/j.trc.2010.10.004.
8.
Özuysal M., S.P. Çalışkanelli, S.
Tanyel, T. Baran. 2004. “Capacity prediction for traffic circles:
applicability of ANN”. Proceedings
of the Institution of Civil Engineers – Transport 162(4): 195-206.
ISSN: 0965-092X. E-ISSN: 1751-7710. DOI:
https://doi.org/10.1680/tran.2009.162.4.195.
10. Patnaik Ashish Kumar,
Yadu Krishna, Shweta Rao, Prashant Kumar Bhuyan. 2017. “Development of
Roundabout Entry Capacity Model Using INAGA Method for Heterogeneous Traffic
Flow Conditions”. Arab J Sci Eng
42(9): 4181-4199. DOI: https://doi.org/10.1007/s13369-017-2677-x.
11. Mauro Raffaele. 2010.
Calculation of Roundabouts: Capacity,
Waiting Phenomena and Reliability. Berlin, Heidelberg: Springer Berlin.
ISBN 978-3-642-04550-9.
12. Ghanizadeh Ali Reza,
Nasrin Heidarabadizadeh, Farhang Jalali. 2020. “Artificial neural network
back-calculation of flexible pavements with sensitivity analysis using Garson’s
and connection weights algorithms”. Innov.
Infrastruct. Solut. 5(2): 63. DOI:
https://doi.org/10.1007/s41062-020-00312-z.
Appendix 1
Geometric specifications of selected unsignalized
roundabouts
|
Sl.
No. |
Site |
City |
Leg At |
Approach width ( |
Entry width |
Weaving width ( |
Weaving length ( |
Diameter of central
island ( |
|
1 |
Sector-2 Square |
Rourkela, Odisha |
E |
9.58 |
11 |
12.5 |
34.02 |
23.83 |
|
W |
18.73 |
13.44 |
28.5 |
43.82 |
||||
|
N |
22.78 |
12 |
21.09 |
36.74 |
||||
|
S |
18.4 |
15.75 |
29.56 |
40.91 |
||||
|
2 |
SAIL Square |
Rourkela, Odisha |
E |
6.53 |
19.36 |
32 |
49.31 |
50 |
|
W |
17.51 |
14.17 |
20.98 |
43.84 |
||||
|
N |
22 |
19.74 |
32.45 |
48.54 |
||||
|
S |
19.92 |
18.36 |
26.05 |
44.27 |
||||
|
3 |
Ambagan Square |
Rourkela, Odisha |
E |
11.24 |
15.1 |
27.19 |
43.99 |
38 |
|
W |
15.14 |
19.81 |
35.01 |
50.22 |
||||
|
N |
23.4 |
11.59 |
23.43 |
41.59 |
||||
|
S |
18.91 |
18.23 |
33.72 |
53.77 |
||||
|
4 |
Plant Side Square |
Rourkela, Odisha |
E |
6.2 |
15.43 |
31.12 |
43.31 |
48.87 |
|
W |
8.1 |
13.68 |
24.25 |
41 |
||||
|
N |
18.01 |
19.17 |
30.96 |
45 |
||||
|
S |
19.2 |
16.12 |
29.58 |
45.07 |
||||
|
5 |
Traffic Gate Square |
Rourkela, Odisha |
E |
12.67 |
15.52 |
23.31 |
43.79 |
43.55 |
|
W |
8.63 |
5.41 |
19.63 |
39.87 |
||||
|
N |
21.27 |
15.74 |
32.25 |
51.7 |
||||
|
S |
20.13 |
8.5 |
21.58 |
37.92 |
|
6 |
Birsa Square |
Rourkela, Odisha |
N |
10.58 |
17.55 |
33.2 |
50.95 |
60.12 |
|
N-E |
9.63 |
16.61 |
32.72 |
52.33 |
||||
|
E |
12.06 |
18.9 |
33.91 |
58.72 |
||||
|
W |
12.77 |
17.22 |
28 |
47.67 |
||||
|
7 |
Panposh Square |
Rourkela, Odisha |
E |
16.07 |
14.3 |
27.9 |
48.21 |
30.28 |
|
S |
16.95 |
13.76 |
23.75 |
42.87 |
||||
|
W |
14.02 |
15.08 |
29.56 |
41.86 |
||||
|
8 |
Ainthapalli Square |
Sambalpur, Odisha |
N-E |
8.01 |
12.93 |
27.57 |
35.56 |
47.88 |
|
N-W |
14.18 |
13.2 |
19.57 |
43.18 |
||||
|
S-E |
13.12 |
18.97 |
30.95 |
46.97 |
||||
|
S-W |
11.66 |
13.33 |
24.85 |
37.95 |
||||
|
9 |
Master Canteen Square |
Bhubaneswar, Odisha |
N |
25.99 |
17.72 |
26.78 |
48.65 |
45.91 |
|
S |
25.56 |
14.55 |
28.99 |
53.44 |
||||
|
E |
16.85 |
14.8 |
32.26 |
49.71 |
||||
|
W |
12.8 |
16.79 |
27.6 |
44.23 |
||||
|
10 |
Gopabandhu Square |
Bhubaneswar, Odisha |
N |
10.1 |
18.33 |
32.56 |
49.83 |
51.62 |
|
E |
10.84 |
17.68 |
29.07 |
47.35 |
||||
|
W |
10.01 |
17.82 |
23.99 |
46.58 |
||||
|
11 |
Jobra Square |
Cuttack, Odisha |
N |
10.2 |
19.68 |
31.25 |
48.71 |
37.21 |
|
S |
8.2 |
16.5 |
27 |
46.71 |
||||
|
N-E |
10.73 |
15.42 |
20 |
41.9 |
||||
|
W |
15.77 |
15.04 |
29.87 |
46.57 |
||||
|
12 |
Palbani Square |
Baripada, Odisha |
E |
9.47 |
18.98 |
30.57 |
47.87 |
|
|
W |
9.4 |
17.96 |
31.35 |
52.3 |
58.88 |
|||
|
N |
9.53 |
16.47 |
27.94 |
50.15 |
|
|||
|
S |
10.45 |
12.41 |
21.44 |
40.86 |
|
|||
|
13 |
Dargadhi Square |
Baripada, Odisha |
E |
10.42 |
14.98 |
23.6 |
45 |
|
|
W |
11.86 |
18.97 |
31.33 |
52.76 |
39.38 |
|||
|
N |
10.89 |
18.12 |
32.7 |
53.28 |
|
|||
|
S |
12.31 |
19.02 |
33.45 |
50.64 |
|
|||
|
14 |
Salt-lake Square |
Kolkata, West Bengal |
NE |
8.85 |
15.85 |
30.1 |
46.08 |
|
|
NW |
8.85 |
11.52 |
21.96 |
42.08 |
33.55 |
|||
|
SE |
8.85 |
13.57 |
22 |
35.09 |
|
|||
|
SW |
8.85 |
11.85 |
28.32 |
36.08 |
|
|||
|
15 |
Albert Ekka Square |
Ranchi, Jharkhand |
N-E |
15.45 |
4.72 |
9.48 |
32.08 |
|
|
S |
19.58 |
5.2 |
8.2 |
29.42 |
10.76 |
|||
|
N-W |
13.38 |
6.31 |
15.39 |
31.57 |
|
|||
|
16 |
Old Bus Stand Square |
Bilaspur, Chhattisgarh |
E |
7 |
18.19 |
34.86 |
51.25 |
|
|
W |
7.2 |
18.25 |
27.35 |
48.7 |
|
|||
|
N |
9.46 |
4.43 |
15.21 |
30.15 |
55 |
|||
|
S |
9.81 |
16.9 |
30.15 |
56.2 |
|
|||
|
17 |
Ramnagar Square |
Nagpur, Maharashtra |
NE |
13.93 |
11 |
23.85 |
40.31 |
|
|
NW |
11.52 |
13.94 |
23.57 |
36.66 |
|
|||
|
W |
11.26 |
11.31 |
25.47 |
37.83 |
|
|||
|
SW |
13.29 |
11.22 |
23.94 |
36.4 |
36.54 |
|||
|
E |
6.93 |
13.04 |
22.33 |
38.12 |
|
|||
|
SE |
15.33 |
5.9 |
14.8 |
31.2 |
|
|||
|
S |
11.45 |
5.12 |
11.02 |
28.96 |
|
|||
|
18 |
Medical Square |
Nagpur, Maharashtra |
N |
6.74 |
6.14 |
13.39 |
38.96 |
|
|
NE |
7.32 |
9.5 |
17.45 |
38.52 |
|
|||
|
SE |
6.38 |
18.36 |
27.01 |
43.8 |
|
|||
|
S |
10.97 |
16.71 |
30.02 |
47.63 |
51.22 |
|||
|
SW |
9.62 |
16.77 |
28.47 |
49.22 |
|
|||
|
NW |
6.71 |
13.97 |
33.02 |
41.15 |
|
|||
|
19 |
Chacka
Junction |
Thiruvananthapuram, Kerala |
NE |
7.64 |
16.34 |
32.01 |
44.32 |
|
|
NW |
6.21 |
14.71 |
26.025 |
45.34 |
46.8 |
|||
|
SE |
8.43 |
16.94 |
26.98 |
39.8 |
|
|||
|
SW |
6.98 |
15.94 |
30.01 |
44.24 |
|
|||
|
20 |
Womens College, Vazhuthakadu |
Thiruvananthapuram, Kerala |
N |
18.39 |
19.1 |
32.67 |
46.85 |
|
|
S |
17.59 |
16.59 |
20.5 |
40.42 |
34.28 |
|||
|
E |
6.77 |
13.85 |
28.8 |
43.45 |
|
|||
|
W |
13.88 |
16.47 |
25.26 |
40.89 |
|
|||
|
21 |
MVP Colony Square |
Visakhapatnam, Andhra Pradesh |
E |
8.12 |
17.3 |
28.9 |
47.98 |
|
|
W |
8.19 |
18.79 |
29.98 |
50.05 |
45.47 |
|||
|
N |
8.09 |
14.65 |
27.59 |
43.1 |
|
|||
|
S |
7.98 |
20.12 |
29.6 |
52.3 |
|
|||
|
22 |
Diamond Park Junction |
Visakhapatnam, Andhra Pradesh |
E |
14.51 |
19.8 |
28 |
53.29 |
|
|
W |
14.01 |
15.81 |
31.65 |
47.13 |
54.8 |
|||
|
N |
18.19 |
18.5 |
25.14 |
47.79 |
|
|||
|
S |
15.52 |
16.88 |
22.51 |
38 |
|
|||
|
23 |
BR Ambedkar Square |
Visakhapatnam, Andhra Pradesh |
E |
11.2 |
18.93 |
30.02 |
45.69 |
|
|
W |
11.71 |
15.3 |
22.91 |
38.31 |
40.1 |
|||
|
N |
10.04 |
19.07 |
29.95 |
41.22 |
|
|||
|
S |
11.05 |
14.79 |
24.9 |
41.02 |
|
|||
|
24 |
Dornama Raju Square |
Visakhapatnam, Andhra Pradesh |
N |
14.29 |
12.15 |
31.84 |
41.59 |
|
|
S |
20.85 |
14 |
29.84 |
41.22 |
40.65 |
|||
|
E |
15.86 |
14.58 |
22.9 |
40.26 |
|
|||
|
W |
20.47 |
10.12 |
22.86 |
47.86 |
|
|||
|
25 |
Sector 43-44 Junction |
Chandigarh |
NE |
13.97 |
16.89 |
33.17 |
44.72 |
|
|
NW |
13.75 |
19.46 |
33.25 |
45.88 |
|
|||
|
SE |
11.68 |
18.48 |
30.04 |
50.89 |
50 |
|||
|
SW |
11.38 |
16.88 |
30.58 |
41.81 |
|
|||
|
26 |
Sector 42 Junction |
Chandigarh |
NE |
8.28 |
19.79 |
29.55 |
51.59 |
|
|
NW |
9.39 |
19.14 |
29.72 |
45.14 |
49 |
|||
|
SE |
8.14 |
19.87 |
29.56 |
46.13 |
|
|||
|
SW |
8.38 |
14.98 |
27.9 |
48.62 |
|
|||
|
27 |
Sector 49 Junction |
Chandigarh |
N |
10.4 |
16.9 |
34 |
56 |
|
|
S |
10 |
16.4 |
30.5 |
48.3 |
57 |
|||
|
E |
10.6 |
19.3 |
31 |
51.6 |
|
|||
|
W |
11 |
16.2 |
32.8 |
52.5 |
|
Note:
N = North, S = South, E = East, W = West, NE
= North-East, NW = North-West, SE = South-East, SW = South- West
Received 06.12.2023; accepted in
revised form 29.03.2024
Scientific Journal of Silesian University of Technology. Series
Transport is licensed under a Creative Commons Attribution 4.0
International License
[1]
Department of Civil and Environmental Engineering, Research Scholar, Birla
Institute of Technology Mesra, Ranchi 835215, Jharkhand, India. Email:
phdcee10051.20@bitmesra.ac.in. ORCID: https://orcid.org/0009-0003-5808-8287