- Email: [email protected]
Analyzing the Impact of Anticipatory Vehicle Routing on the Network Performance
Available online at www.sciencedirect.com Available online at www.sciencedirect.com
ScienceDirect ScienceDirect
Transportation Research Procedia 00 (2018) 000–000 Available online at www.sciencedirect.com Transportation Research Procedia 00 (2018) 000–000
ScienceDirect
www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia
Transportation Research Procedia 41 (2019) 494–506
mobil.TUM 2018 "Urban Mobility – Shaping the Future Together" - International Scientific mobil.TUM 2018 "Urban Mobility – Shaping the Future Together" - International Scientific Conference on Mobility and Transport Conference on Mobility and Transport
Analyzing the Impact of Anticipatory Vehicle Routing on the Analyzing the Impact of Anticipatory Vehicle Routing on the Network Performance Network Performance a, a b b Aledia Bilali *, Gordon Isaac , Sasan Amini , Nassim Motamedidehkordi Aledia Bilalia,*, Gordon Isaaca, Sasan Aminib, Nassim Motamedidehkordib a BMW Group, Parkring 19, Garching 85748, Germany Technischea BMW Universität München 80333, Germany Group,München, ParkringArcisstrasse 19, Garching21,85748, Germany b Technische Universität München, Arcisstrasse 21, München 80333, Germany b
Abstract Abstract Traffic congestion causes lost in time, fuel consumption and frustration of the drivers. Route guidance systems are considered as the mostcongestion feasible solution alleviate congestion. However, existing navigation systems provide route based Traffic causes to lost in time,traffic fuel consumption and frustration of route the drivers. Route guidance systems areguidance considered as only on real-time traffic information, consideringHowever, the futureexisting evolvement traffic insystems the network. These strategies, the most feasible solution to alleviate without traffic congestion. route of navigation provide route guidanceknown based as reactive routingtraffic strategies, react only whenconsidering congestionthe hasfuture already occurred of and do not prevent congestion from happening. only on real-time information, without evolvement traffic in the network. These strategies, known Therefore, have been in anticipatory guidance, enable the prevent integration of trafficfrom prediction into as reactive researchers routing strategies, reactinvestigating only when congestion has route already occurredtoand do not congestion happening. route recommendation. The anticipatory route guidance is expected improvetothe effectiveness of the of route guidance systems, Therefore, researchers have been investigating in anticipatory route to guidance, enable the integration traffic prediction into however its effects on the traffic on theguidance network is areexpected unclear. to The main aim this paper isoftothe analyze the impacts of the route recommendation. Theoverall anticipatory route improve the of effectiveness route guidance systems, anticipatory routingon strategy on the network performance. achieve theaim performance of istheto system anticipatory however its effects the overall traffic on the network are To unclear. Thethis, main of this paper analyzeduring the impacts of the routing strategy is compared to current reactive routing strategy. The models the routing strategies are built andanticipatory tested in a anticipatory routing strategy on the network performance. To achieve this, theofperformance of the system during microscopic simulation environment andreactive the results are compared suitableofnetwork performance indicators, as average routing strategy is compared to current routing strategy. using The models the routing strategies are built such and tested in a travel time, total delay environment time and the and effects of anticipatory routing on suitable differentnetwork vehicle performance types. Evaluation resultssuch show that the microscopic simulation the results are compared using indicators, as average anticipatory routing strategy the network performance, for a limited rate of the vehicles travel time, total delay time can andimprove the effects of anticipatory routingbut ononly different vehicle penetration types. Evaluation results showequipped that the with navigation system basedcan on improve the predictive traffic information, timepenetration compared to reactive reduced anticipatory routing strategy the network performance,the butaverage only fortravel a limited rate of the routing vehiclesisequipped by upnavigation to 20 % insystem the best scenario the vehicles with the predictive navigation system the ones profit routing the mostisfrom the with based on theand predictive traffic information, the average travel time are compared to to reactive reduced predictive strategy. by up to 20routing % in the best scenario and the vehicles with the predictive navigation system are the ones to profit the most from the predictive routing strategy. © 2018 The Authors. Published by Elsevier B.V. © 2019 The Authors. Published by Elsevier Ltd. Peer-review under responsibility of Elsevier the of the mobil.TUM 2018 conference. © 2018 The Authors. Published by B.V. committee This is an open access article under the scientific CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the mobil.TUM 2018 conference. Peer-review under responsibility of the scientific committee of the mobil.TUM18. Keywords: Anticipatory routing; traffic prediction; network performance Keywords: Anticipatory routing; traffic prediction; network performance
* Corresponding author. Tel.: +491736772042. E-mail address:author. [email protected] * Corresponding Tel.: +491736772042. E-mail address: [email protected] 2352-1465 © 2018 The Authors. Published by Elsevier B.V. Peer-review the scientific committee 2352-1465 ©under 2018responsibility The Authors. of Published by Elsevier B.V.of the mobil.TUM 2018 conference. Peer-review under responsibility of the scientific committee of the mobil.TUM 2018 conference.
2352-1465 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the mobil.TUM18. 10.1016/j.trpro.2019.09.082
2
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
495
1. Introduction Urban road networks are experiencing a higher level of traffic and long wasted time in traffic jams every day. Besides the increased level of pollution, higher fuel consumption and large economic losses, traffic congestion causes also mental stress and frustration to the drivers (Eisele et al., 2012). With the rapidly growing population and urbanization, traffic congestions are estimated to worsen in the future. Congestion happens when traffic demand exceeds network capacity. Therefore, a common approach previously used to mitigate congestion was to construct new roads or widen the existing ones, in order to increase the capacity to easily accommodate the increased demand. However, construction of new roads, especially in urban areas, where higher amount of congestion is registered, is firstly costly and it is not sure if it might mitigate congestion, considering the induced traffic that will be caused due to the new constructions. Nowadays three solutions are considered to be able to alleviate congestion: the reduction of travel demand, shifting towards other modes of transport or distributing traffic to better utilize the network capacity (Liang and Wakahara, 2014). As with the increasing population, the number of trips is projected to be increased and not reduced and influencing the drivers to shift to other modes is difficult, using route guidance systems to distribute traffic is seen as the most feasible solution to improve urban traffic jams. The integration of new technologies which provide access to traffic data and the available computational power are major contributors to the successful implementation of route guidance systems, which support congestion mitigation. Intelligent Transportation System attempts to use route guidance systems or adaptive traffic lights to reduce the effects of congestion, by accessing traffic data, via inductive loops, mobile devices or new technologies, such as image processing. Moreover, with the advancement in communication technologies, such as the use of wireless communication, a car has evolved to a sensor platform (Kim et al., 2016). Having greater access to the infrastructure data and more information about the state of the network and possible congestion, vehicle route navigation systems can compute the shortest travel time path based on real-time traffic information, rather than the calculation of the shortest path that was available in the past. Nevertheless, existing route navigation systems provide route guidance based only on real-time traffic information, without considering the future evolvement of traffic in the network. These strategies, known as reactive routing strategies, react only when congestion has already occurred and do to prevent congestion from happening. The future deployment of anticipatory routing strategies, which consider anticipated traffic condition, would probably enhance the effectiveness of the in-car navigation systems, especially in large scale dynamic environments. However, the effects that this routing strategy will have on the network performance, if the route recommendation is provided to a certain number of drivers, are not yet clear. The main goal of this study is to investigate the impacts of anticipatory routing strategy compared to current reactive routing strategy on the network performance, for different penetration rates of vehicles equipped with the predictive navigation device. To compare the results of different routing strategies, the main performance measures that will be used are average travel time and total delay time. Moreover, the impacts of anticipatory routing strategy on different drivers will also be investigated by evaluating the average travel time of different vehicle types. The rest of the paper is organized as follows. Section 2 describes the related work on the topic. Section 3 and Section 4 explain the system model and the simulation setup used to access the results. In Section 5 the evaluation and discussion of the results is performed. Conclusions are described in Section 6. 2. Related Work A lot of research is performed in anticipatory routing and its effects on the network performance. The researchers stress the benefits of anticipatory routing compared to reactive routing mainly in terms of travel time and congestion avoidance, especially when the penetration rate of the vehicles equipped with a navigation system based on the predictive traffic information is high. Following is an overview of the main research works performed on this topic. Kim et al. (2016) in their study compare Dynamic routing with prediction with Static and Dynamic routing. In Static routing the vehicles follow the routes that were assigned to them at the beginning without any change, while in Dynamic routing the route is updated depending on real-time traffic information. Their results indicate that the Dynamic with prediction reduces the travel time effectively compared with the other strategies, especially when the
496
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
3
penetration rate of the vehicles with the predictive navigation device is high. However, after rerouting the first vehicle, they penalize the route that the first vehicle has already chosen, when calculating the route for the second vehicle in the next time step. This is not the case in reality, as it is difficult to take into account the impact of the rerouted vehicle in the network immediately when route guidance is given to the next vehicle. Pan et al. (2013) also studied the benefits of anticipatory routing. They developed five rerouting strategies which proactively compute individually-tailored route guidance to vehicles when there are signs of congestion in the network. Their simulation results show that these strategies can reduce travel time as much as the state of art dynamic traffic assignment (DTA), while requiring less computational time. However, they estimate travel time based on speeddensity relationship using Greenshields’s model, which was initially developed for motorways and does not necessarily hold in urban traffic. To compensate the drawbacks of the modelling approaches which use speed-density relation for traffic prediction in the urban environment, Liang and Wakahara (2014) adopted a microscopic modelling and developed two different prediction methods. The first prediction model was based on spatiotemporal correlation of the actual link and all the other upstream links. This model uses adjustment factors which take into account the effects of traffic lights on traffic dynamics and assumes that the speed is taken from loop detectors rather than derived from conventional speed-density relation. Their second prediction model is based on the approach that the inflow and the outflow on the concerned link are based on the spare load capacity. The results show the benefit of anticipatory routing leading to 70% reduction in average travel time compared to reactive routing. Claes et al. (2011) use a decentralized approach for anticipatory vehicle routing using delegate multi-agent system in which vehicles are routed based on current and predictive traffic information. Their routing approach was inspired by the food foraging mechanism in ant colonies, using pheromones. Multi-agent collect information in the network and distribute it to a central server which computes the best path for a given origin-destination (OD) pair. The experiments show that this strategy helps the drivers who have the predictive traffic information to reach their destination 35 % faster compared to the drivers who do not have data or have real-time traffic data. The approach used in this paper is different from the before mentioned approaches because firstly, it uses a microscopic simulation to estimate the travel time in the network. Moreover, it considers the current procedure of incar navigation systems, when depending on traffic information at a certain point in time, the same route guidance is given to all the vehicles equipped with a navigation device. This technique represents the reality better in terms of vehicle distribution based on a route guidance system, because contrary to the other approaches it does not consider each previously rerouted vehicle one by one in the route recommendation to the next vehicles, which would be the case in an ideal situation. 3. System model 3.1. Basic model A simple test network is built with the help of a microscopic traffic simulation to evaluate the two different routing strategies. DTA is chosen as an assignment procedure as it is more realistic compared to static traffic assignment, allows modelling of variations on travel demand and avoids entering the routes manually (PTV GROUP, 2016; Chiu et al., 2011). Traffic assignment is essentially a model of the route choice in general and the model of this decision is a special case of the discrete choice modelling (Fellendorf and Vortisch, 2010). DTA represents an iterative learning process and calculates the best path for each iteration taking into account the experience from the previous iterations. With the help of DTA, the dynamic user equilibrium can be reached, which means that vehicles departing at a specific point in time among the same OD pair, but choosing different routes, have the same experienced travel time (Chiu et al., 2011). This section describes how the dynamic assignment is performed in this study to achieve a realistic vehicle distribution in the network in the lack of real travel demand data. 3.1.1. Test network and travel time calculation The network tested is represented by nodes and edges. In this representation, intersections are modelled as nodes and roads between the intersections as edges of the network graph. Accordingly, the network area is divided into
4
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
497
subareas and the number of trips between the subareas for a given time interval are defined by a matrix. The travel time is measured for each edge in the network graph. The vehicles report the travel time when leaving the edge. All the travel times reported from the vehicles for an evaluation interval are averaged and the travel time for the edge is calculated. In case of congestion, vehicles also report that they couldn’t leave the edge during that evaluation interval, by reporting their dwell time as well, allowing in this way the detection of heavily congested edges (PTV GROUP, 2016). The travel time of the current iteration is used for path selection for the next iteration considering also the travel time from more distant measurements. The method of exponential smoothing of travel times, with a smoothing factor of 0.5 is used to give 50 % influence to the most recent iteration and less influence to the earlier ones. The smoothed travel time is calculated using the following formula (Fellendorf and Vortisch, 2010): ����� � �� � ������ � � �1 � �� � �������
(1)
Where: � : index of the evaluation interval within the simulation time � : index of the iteration � : index of the edge ������ : measured (observed) travel time on edge � for the interval � in the iteration � ����� : smoothed travel time edge � for the interval � in the iteration � � : the given constant smoothing factor 3.1.2. Path selection
In an abstract graph a path is considered to be a sequence of edges. Therefore, the generalized costs of a path are the sum of the generalized costs of the edges that form that path. In DTA procedure, the best path for an OD trip is found iteratively (Chiu et al., 2011). Firstly, the best path is calculated based on the distance to the destination. In the other iterations, all the available paths to the destination are investigated by giving a zero weight to the paths which were not checked. After all the paths are checked, for each iteration the traffic situation and the travel time on the edges change. Therefore, different best paths are selected until convergence is reached. For the distribution of the travel demand of OD pairs, the Kirchhoff distribution formula is used (Fellendorf and Vortisch, 2010): ���� � � �
���
(2)
∑� ���
Where: �� : the benefit of path � ���� � : the probability that path � is selected � : the sensitivity parameter of the model 3.1.3. Vehicle types
In this study, four different vehicle types are used to represent different navigation systems that the vehicles are equipped with and the usage of these navigation systems. The vehicle types and their general cost functions are described below: 1.
Experienced drivers – normal in-city drivers who do not use navigation system and choose the route only based on experience
��������������1� � �1� � ����������������������
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
498
2.
3. 4.
5
Vehicles with distance navigation system (DIST-NS) – drivers who choose the route that seems to be the shortest depending only on free flow conditions ������� ���� ��� � � � ������ �������� Vehicles with real-time traffic information navigation system (RTTI-NS)
������� ���� ��� � � � ������� ������ ���� Vehicles with predictive traffic information navigation system (PREDTI-NS)
������� ���� �4� � � � ���������� ������ ����
In order to create the basic network setting, initially only the first two types of the vehicles (Experienced drivers and vehicles with DIST-NS) are used. In the next phases, the other two types of vehicles are also considered. 3.2. Reactive routing strategy model This strategy offers route guidance based on real-time traffic information and is the current strategy used to provide route guidance to the vehicles. Fig. 1 describes the procedure used to model this type of routing strategy. After iterative simulations using DTA, the conditions of how the traffic is distributed in reality are assumed to be reached. Then in order to give route guidance to the vehicles with RTTI-NS, static routes is created from DTA. As the network is represented by a node-edge graph, travel time of the edges is needed to be known in order to incorporate it in the algorithm which finds the shortest travel time path.
Fig. 1. Reactive routing strategy model
Travel time for each edge is calculated based on the speed of the edges. During the simulation, the speed for each edge is derived and aggregated in five minutes intervals until the end of the simulation run. The interval of five minutes is chosen because traffic flow and speed oscillate too much on smaller intervals (Guo et al., 2008). Based on the speed and link distance, travel time for each edge in the network is calculated every 5 minutes based on the exact physical relationship:
���� �
� �
Where: t ��� : current travel time d : link distance � : speed on the link
( 3 )
As the network tested is a small network and computation time is not a concern, Dijkstra algorithm (Dijkstra, 1959) is selected as a routing algorithm. Dijkstra algorithm uses the aggregated travel time to find the shortest travel time
6
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
499
path, every five minutes. Then the route guidance is given to the vehicles which are equipped with RTTI-NS. The procedure is repeated for every time interval of five minutes until the end of the simulation. In order to represent the delay in processing traffic information, the shortest route is given to the vehicles five minutes after the shortest travel time path is calculated. For the specific penetration rate of vehicles equipped with RTTI-NS that is selected, a compliance rate of 100% is assumed. 3.3. Anticipatory routing strategy model In anticipatory routing, traffic prediction is integrated into the route guidance system. The predictive travel time is derived using microscopic simulation as a prediction method. The shortest travel time path based on this predictive travel time is then calculated by Dijkstra algorithm and the route guidance is given to the vehicle equipped with the PREDTI-NS. Additionally, route guidance based on current traffic information is provided to the vehicles with RTTINS. 3.3.1. Traffic prediction In order to obtain the predicted travel time for the edges, the parametric prediction method based on microscopic traffic simulation is used. Obtaining the predicted travel time from simulations is meaningful especially in urban environments, due to consideration of the delay caused by the presence of traffic lights and other disturbing traffic. In the case of an incident, as the one tested in this study, using traffic simulation as a prediction tool is assumed to give more accurate traffic predictions results compared to the other prediction methods (Van Lint and Van Hinsbergen, 2012). In this study, predicted travel time data are provided by running the simulation in advance. The simulation that is run in advance, has the same settings as the previously described model of reactive routing strategy. This model, apart from being used as a comparison model, is also used as a basis for the prediction. Fig. 2 illustrates how the simulation of reactive routing model is used as a basis for travel time prediction in anticipatory routing model. During the simulation of reactive routing strategy model, the travel time for each edge is calculated every five minutes, as described in reactive routing section. The traffic information containing the travel time of each edge in the network, every five minutes until the end of the simulation period, are saved. The predicted travel time is derived from the before mentioned saved travel time data five minutes in advance. For instance (Fig. 2), travel time data for each edge in the network, collected during reactive routing model at the point in time ‘ ݐ ͷǯ, is used as predicted travel time data for the edges in the anticipatory routing model at the point in time ‘’ݐ.
Fig. 2. Travel time prediction
Following this procedure, predicted travel time data, showing the state of the travel time five minutes in advance for each edge is available for the anticipatory routing model every five minutes, by using simulation as a prediction tool. 3.3.2. Anticipatory routing model The anticipatory route model considers the predictive traffic information to give the route guidance. As described
500
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
7
in the previous section the predictive travel time is used to calculate the shortest path using Dijkstra routing algorithm and give the guidance to the vehicles. In this routing strategy (Fig. 3), the guidance is given to the vehicles with RTTI-NS and vehicles with PREDTINS simultaneously. The difference is that the vehicles equipped with the predictive navigation system obtain the route guidance based on predictive traffic information and the vehicles with the real-time navigation system obtain the route guidance based on current traffic information.
Fig. 3. Anticipatory routing model
4. Simulation setup The network tested represents an urban grid network, with a surface of around 9 km2 (Fig. 4). The length of the links varies from 200 m to 900 m. All links are two ways links, with one lane per direction. The intersections are all equipped with traffic lights which have a fixed signal program. Traffic demand is generated in the network from four origins and distributed in the network based on four OD pairs. In order to create the initial network setup a realistic distribution of traffic in the network is needed, therefore, as already described, the dynamic traffic assignment is selected as a traffic assignment method.
Fig. 4. Urban network topology in VISSIM
8
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
501
Modeling and testing of the routing strategies are performed in the traffic microsimulation environment VISSIM. Python programing language in combination with COM interface is used to integrate Dijkstra routing algorithm in VISSIM and give the route guidance to the vehicles with RTTI-NS and PREDTI-NS. For a better evaluation of the significance of the results, the anticipatory routing strategy is compared with reactive and fixed path routing strategy. A set of different scenarios considering different penetration rate of different vehicle types are considered for further evaluation. A summary of different scenarios which are tested is shown in Table 1. Scenario F refers to the fixed path routing strategy. In this scenario, the vehicles paths are fixed and do not change when the traffic conditions change. Two vehicle types are present: the experienced drivers and vehicles with DISTNS. Experienced drivers comprise 70 % of the vehicles and choose the route based on experience. Whereas, the vehicles with DIST-NS, which are assumed to take two best routes based on free flow conditions, include 30 % of the vehicles. Scenario R belongs to the reactive routing strategy. Despite the experienced drivers and the vehicles with DIST-NS, the vehicles which have RTTI-NS are also part of the network system in this scenario. These vehicles, which comprise 20 % of the total vehicles, have an updated path every five minutes, based on real-time traffic information. The percentage of experienced drivers and vehicles with DIST-NS is 50 % and 30 %, respectively. Scenario A comprises all the scenarios based on anticipatory routing strategy. In addition to experienced drivers, vehicles with DIST-NS and RTTI-NS, vehicles equipped with PREDTI-NS are also present. The shortest travel time path for this type of vehicles is calculated based on the predictive travel time and their route is updated every five minutes. Scenario A1 until Scenario A6 refer to different penetration rate of this vehicle type, starting from 5 % until 40 % infiltration in the network. The drivers with a real-time navigation and predictive navigation are assumed to comply 100% with the route guidance given from their navigation devices. Table 1. Percentage of different vehicle types for different scenarios. Routing Strategy ___________ Vehicle Type Experienced drivers
Fixed path routing strategy
Reactive routing strategy
70 %
50 %
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
5 % Pred.
10 % Pred.
20 % Pred.
25 % Pred.
30 % Pred.
40 % Pred.
45 %
40 %
30 %
25 %
20 %
20 %
DIST-NS
30 %
30 %
30 %
30 %
30 %
30 %
30 %
20 %
RTTI-NS
0%
20 %
20 %
20 %
20 %
20 %
20 %
20 %
PREDTI-NS
0%
0%
5%
10 %
20 %
25 %
30 %
40 %
Scenarios
F
R
A1
A2
A3
A4
A5
A6
5. Evaluation Based on the indicators of mobility assessment for network performance defined by TRANSPORTATION ASSOCIATION OF CANADA (2006), for the purpose of this study, mean travel time and total delay time, are selected as the main network performance indicators. Moreover, the average travel time of specific vehicle types is also considered in the evaluation. To calculate the average travel time for all vehicles, a weighted average is selected (Madansky, 2017), due to the fact that the average travel time for each evaluation interval is dependent on the number of vehicles for the corresponding interval. The total delay time is obtained by subtracting the actual travel time in a specific time step with the travel time when traveling with the desired speed. An overview of the results for each indicator is presented in this section. The evaluation of the average travel time and the delay time is performed under changing demand conditions, representing a rush hour period and changing capacity conditions, where an incident blocking one lane for 20 minutes is modelled. The travel time was determined for four different OD pairs. The incident that was modelled influenced directly the vehicles following OD 1, partly vehicles following OD 4, and did not directly influence the vehicles following the other two OD pairs.
502
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
9
To capture all these changing conditions the network is simulated for five hours. The evaluation period does not include 30 minutes of warm-up time and 15 minutes of warm-end time. Furthermore, the evaluation is performed separately for the incident period, because the impacts of the rerouting strategies are more obvious during this time frame. The results considered for the analyses were clustered in five minutes intervals. For the purpose of this study, a minimum number of five simulation runs with different random seeds was required for a confidence level of 95 %, based on the procedure described by Dowling et al. (2004). The values from all simulation runs were averaged and were taken into account for further analyses. 5.1. Average travel time Fig. 5 presents the average travel time for all vehicle types during the incident period. During normal free flow conditions, it takes around 12 minutes to perform the trip for each OD pair. However, during the incident period the travel time for reactive routing strategy is increased up to 19 minutes and 15 minutes for OD 1 and OD 4, respectively. These two OD pairs are directly affected by the incident, with more impacts experienced by the vehicles traveling in OD 1. In the best scenario (Scenario A3) the average travel time of all vehicles in the network is around 15 minutes and 13 minutes for OD 1 and OD 4, respectively, showing lower values compared to reactive routing strategy, where the average travel time is 19 and 15 minutes, respectively. Vehicles travelling in OD 2 and OD 3 are not directly influenced by the incident and their travel time during the incident conditions is almost the same as during free flow conditions, therefore their travel time varies slightly compared with reactive routing. For these two OD pairs, the travel time when using anticipatory routing might be even higher than when using reactive routing, because the travel time in free flow conditions cannot be further improved. Additionally, the vehicles which are rerouted interfere with the vehicles coming from other origins and they cause minor congestion in the locations where no congestion was existent before.
Average travel time [min]
Incident period: Average travel time 25 20 15 10 5 0
OD 1
OD 2
OD 3
OD 4
Scenario F
Scenario R
Scenario A1
Scenario A2
Scenario A3
Scenario A4
Scenario A5
Scenario A6
Fig. 5. Incident period: Average travel time
A slight decrease in travel time is noticed for reactive routing strategy (Scenario R) compared to fixed path routing strategy (Scenario F) during the incident period. This stresses the assumption that reactive routing does not help significantly in travel time reduction. As the results of travel time during fixed path and reactive routing strategy are almost the same, only the reactive routing strategy will be considered for further analyses. Fig. 6 depicts the differences in travel time between reactive and anticipatory routing strategy for different penetration rate of vehicles with PREDTI-NS, represented by different scenarios. For vehicles following OD 1, which suffer the effects of congestion the most, anticipatory routing strategy decreases the average travel time of all vehicle types for all scenarios, except Scenario A6. Until 20 % penetration rate of vehicles with the predictive navigation (Scenario A3), the travel time shows a decreasing trend. For Scenario A3, the highest improvement of 20 % reduction
10
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
503
in average travel time compared to travel time during reactive routing, is reached. When the penetration rate is higher than this level (Scenario A4, A5 and A6) the travel time starts increasing again, however until Scenario A5 it is still 10 % less than the travel time for Scenario R. Contrary to the above-mentioned scenarios, when the travel time of all vehicle types decreased compared to reactive routing strategy, in Scenario A6, the travel time of all vehicles in the network is 5 % higher than the travel time during Scenario R. The reason behind this is because all the vehicles that are rerouted take the same route. Consequently, they avoid congestion at the point where the incident happened, but they are also the reason for causing congestion in the routes where they are led to. The reduction in travel time during reactive routing strategy is obvious for all vehicle types. This happens because rerouting of the vehicles with PREDTINS to other routes improves their individual travel time, but also releases the congested links from traffic demand, improving indirectly the travel time of the other vehicle types. For OD 4 reduction in travel time is observed only for Scenario A1, A2 and A3. The highest improvement for this OD is 9 % in travel time reduction and it is reached for Scenarios A1 and A2.
Average travel time comparison [%]
Average travel time comparison with reactive routing (%) 40 30
22
20
5
10 0 ‐10 ‐20 ‐30
‐11‐12 ‐13‐10 ‐20
3 ‐1 ‐5 ‐6 ‐7 ‐2
OD 1 Scenario A1 Scenario A4
33
30
‐2
OD 2
17
11 7 9
‐9 ‐9 OD 3
Scenario A2 Scenario A5
‐1 0
OD 4 Scenario A3 Scenario A6
Fig. 6. Average travel time comparison with reactive routing
5.2. Average travel time of different vehicle types Fig. 7 shows the average travel time for different vehicle types for OD 1, as a better representative, because all the vehicle types considered in the study use this OD pair. Except vehicles with PREDTI-NS, the average travel time of the other vehicle types showed the same tendency during different anticipatory routing scenarios. A decreasing trend was shown until Scenario A3, then the average travel time for experienced drivers, vehicles with DIST-NS and vehicles with RTTI-NS, was increased again reaching even higher values than during reactive routing for Scenario A6. The highest reduction in average travel time for these vehicle types compared to reactive routing strategy was depicted for Scenario A3. For this scenario the travel time for experienced drivers, vehicles with DIST-NS and vehicles with RTTI-NS was denoted to be 15, 17, 13 minutes in absolute values, respectively. Contrary to other vehicle types, the average travel time for vehicles with PREDTI-NS indicated increasing values when the penetration rate of these vehicles was increasing, concluding that the first drivers who will adopt the new technology and will have a navigation device based on predictive traffic information will be the one profiting the most.
504
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
11
Average travel time [min]
Average travel time of different vehicle types (OD 1) 30 20 10 0
Experienced drivers
DIST‐NS
RTTI‐NS
PREDTI‐NS
Scenario F
Scenario R
Scenario A1
Scenario A2
Scenario A3
Scenario A4
Scenario A5
Scenario A6
Fig. 7. Average travel time of different vehicle types (OD 1)
Comparing the travel time of different vehicle types for a specific anticipatory routing scenario, it was noted that for all the considered scenarios, the vehicles which profit the most from anticipatory routing strategy were the vehicles with PREDTI-NS. Their profit was more considerable for lower penetration rates of this vehicles type, with average travel time close to free flow condition, 12 and 13 minutes, respectively for Scenario A1 and A2, which correspond to 20 % and 13 % shorter travel time than for vehicles with RTTI-NS. The vehicles with RTTI-NS are the second ones to profit from anticipatory routing strategy in terms of travel time reduction, followed by experienced drivers and vehicles with DIST-NS. 5.3. Total delay time The total delay time performance indicator is compared with reactive routing strategy and the results are indicated in Fig. 8. It was observed that until a penetration rate of 25 % for vehicles with PREDTI-NS (Scenario A4), the total delay time compared to reactive routing strategy was decreased. The highest reduction in total delay time of 21.9 % is depicted for Scenario A2, when the penetration rate of the vehicles with PREDTI-NS was 10 %. Although the highest profit in travel time for OD 1, which was directly influenced by the incident, was in Scenario A3, for the total delay time, the scenario is different due to consideration of all OD pairs in the network. For Scenario A3 the interference between the vehicles coming from different directions is higher and it influences the lower result in total delay time profit compared to Scenario A2, where the interference is lower due to lower penetration rate of vehicles with PREDTI-NS.
12
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
505
Total delay comparison [%]
Incident period: Total delay comparison with reactive routing strategy [%] 30,0
24,3
20,0 4,9
10,0 0,0 ‐2,3
‐10,0 ‐20,0 ‐30,0
‐10,8
‐11,0 ‐21,9
Scenario A1Scenario A2Scenario A3Scenario A4Scenario A5Scenario A6 Fig. 8. Incident period: Total delay comparison
As expected, considering travel time results, for Scenario A5 and Scenario A6, the total delay time during anticipatory routing is higher than the total delay time during reactive routing. The reason is that the vehicles with PREDTI-NS, when rerouted to other paths trying to avoid the congestion caused by the incident, transfer the congestion in other parts of the network due to high amount of vehicles following the same path. The results are particularly worsen for Scenario A6, when the penetration rate of vehicles with predictive navigation is 40 %, resulting in 24 % higher total delay time than during reactive routing. 6. Conclusion This study evaluates the impacts of anticipatory routing strategy on the network performance with means of microscopic traffic simulation in a grid urban network, in which the traffic prediction is simulation-based. The anticipatory routing strategy is compared to the current reactive routing strategy in terms of average travel time, total delay time and the effects on different vehicle types. The results indicate that the anticipatory routing strategy can improve the network performance until a certain penetration rate of vehicles equipped with PREDTI-NS. Depending on the penetration rate, the improvement can also be significant compared to current reactive routing strategy. In terms of average travel time reduction, the profit is noticed until a level of 30 % vehicles with PREDTI-NS, with the highest profit of 20 % travel time reduction, recorded when the penetration rate of this vehicle type is 20 %. For higher levels than 30 % of vehicles with PREDTI-NS, anticipatory routing strategy does not improve the average travel time to any further extent due to the high number of rerouted vehicles which relocate the congestion to other parts of the network. In terms of total delay time, the advantage of anticipatory routing compared to reactive routing was shown until a penetration rate of 25 % of vehicles with PREDTI-NS, with the highest improvement of 22 % decrease in total delay time when the penetration level of vehicles with PREDTI-NS was 10 %. Vehicles which profit the most from anticipatory routing strategy are the vehicles equipped with PREDTI-NS, because having predictive traffic information helps them to avoid congested links. To sum up, it would be very beneficial from the network performance perspective, to integrate predictive traffic information in navigation systems to give route guidance to the vehicles, but this profit would be possible only until a certain penetration rate of the vehicles equipped with PREDTI-NS. Moreover, other routing strategies are needed to be integrated into the system to support a better distribution of the vehicles in the network and improve the network performance.
506
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
13
References Chiu, Y.-C., Bottom, Bottom, Jon, Mahut, M., Paz, A., Balakrishna, R., Waller, T. and Hicks, J. (2011), Dynamic Traffic Assignment: A Primer, Transportation Research Circular. Claes, R., Holvoet, T. and Weyns, D. (2011), “A Decentralized Approach for Anticipatory Vehicle Routing Using Delegate Multiagent Systems”, IEEE Transactions on Intelligent Transportation Systems, Vol. 12 No. 2, pp. 364– 373. Dijkstra, E.W. (1959), “A note on two problems in connexion with graphs”, Numerische Mathematik, Vol. 1 No. 1, pp. 269–271. Dowling, R., Skabardonis, A. and Alexiadis, V. (2004), Traffic Analysis Toolbox Volume III: Guidelines for Applying Traffic Microsimulation Modeling Software. Eisele, B., Schrank, D. and Lomax, T. (2012), “TTI’s 2012 URBAN MOBILITY REPORT”. Fellendorf, M. and Vortisch, P. (2010), “Microscopic Traffic Flow Simulator VISSIM”, in Barceló, J. (Ed.), Fundamentals of Traffic Simulation, International Series in Operations Research & Management Science, Vol. 145, Springer New York, New York, NY, pp. 63–93. Guo, J., Williams, B. and Smith, B. (2008), “Data Collection Time Intervals for Stochastic Short-Term Traffic Flow Forecasting”, Transportation Research Record: Journal of the Transportation Research Board, Vol. 2024, pp. 18–26. Kim, K., Kwon, M., Park, J. and Eun, Y. (2016), “Dynamic Vehicular Route Guidance Using Traffic Prediction Information”, Mobile Information Systems, Vol. 2016, pp. 1–11. Liang, Z. and Wakahara, Y. (2014), “Real-time urban traffic amount prediction models for dynamic route guidance systems”, EURASIP Journal on Wireless Communications and Networking, Vol. 2014 No. 1, p. 85. Madansky, A. (2017), Weighted Standard Error and its Impact on Significance Testing, available at: http://www.analyticalgroup.com/download/WEIGHTED_MEAN.pdf (accessed 2 May 2017). Pan, J., Popa, I.S., Zeitouni, K. and Borcea, C. (2013), “Proactive Vehicular Traffic Rerouting for Lower Travel Time”, IEEE Transactions on Vehicular Technology, Vol. 62 No. 8, pp. 3551–3568. PTV GROUP (2016), PTV Vissim 8 User Manual. Transportation Association of Canada (2006), Performance Measures for Road Networks: A Survey of Canadian Use. Van Lint, H. and Van Hinsbergen, C. (2012), Artificial Intelligence Applications to Critical Transportation Issues: Short-Term Traffic and Travel Time Prediction Models, TRANSPORTATION RESEARCH CIRCULAR, E-C168, Transportation Research Board, Washington, D.C.
ScienceDirect ScienceDirect
Transportation Research Procedia 00 (2018) 000–000 Available online at www.sciencedirect.com Transportation Research Procedia 00 (2018) 000–000
ScienceDirect
www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia
Transportation Research Procedia 41 (2019) 494–506
mobil.TUM 2018 "Urban Mobility – Shaping the Future Together" - International Scientific mobil.TUM 2018 "Urban Mobility – Shaping the Future Together" - International Scientific Conference on Mobility and Transport Conference on Mobility and Transport
Analyzing the Impact of Anticipatory Vehicle Routing on the Analyzing the Impact of Anticipatory Vehicle Routing on the Network Performance Network Performance a, a b b Aledia Bilali *, Gordon Isaac , Sasan Amini , Nassim Motamedidehkordi Aledia Bilalia,*, Gordon Isaaca, Sasan Aminib, Nassim Motamedidehkordib a BMW Group, Parkring 19, Garching 85748, Germany Technischea BMW Universität München 80333, Germany Group,München, ParkringArcisstrasse 19, Garching21,85748, Germany b Technische Universität München, Arcisstrasse 21, München 80333, Germany b
Abstract Abstract Traffic congestion causes lost in time, fuel consumption and frustration of the drivers. Route guidance systems are considered as the mostcongestion feasible solution alleviate congestion. However, existing navigation systems provide route based Traffic causes to lost in time,traffic fuel consumption and frustration of route the drivers. Route guidance systems areguidance considered as only on real-time traffic information, consideringHowever, the futureexisting evolvement traffic insystems the network. These strategies, the most feasible solution to alleviate without traffic congestion. route of navigation provide route guidanceknown based as reactive routingtraffic strategies, react only whenconsidering congestionthe hasfuture already occurred of and do not prevent congestion from happening. only on real-time information, without evolvement traffic in the network. These strategies, known Therefore, have been in anticipatory guidance, enable the prevent integration of trafficfrom prediction into as reactive researchers routing strategies, reactinvestigating only when congestion has route already occurredtoand do not congestion happening. route recommendation. The anticipatory route guidance is expected improvetothe effectiveness of the of route guidance systems, Therefore, researchers have been investigating in anticipatory route to guidance, enable the integration traffic prediction into however its effects on the traffic on theguidance network is areexpected unclear. to The main aim this paper isoftothe analyze the impacts of the route recommendation. Theoverall anticipatory route improve the of effectiveness route guidance systems, anticipatory routingon strategy on the network performance. achieve theaim performance of istheto system anticipatory however its effects the overall traffic on the network are To unclear. Thethis, main of this paper analyzeduring the impacts of the routing strategy is compared to current reactive routing strategy. The models the routing strategies are built andanticipatory tested in a anticipatory routing strategy on the network performance. To achieve this, theofperformance of the system during microscopic simulation environment andreactive the results are compared suitableofnetwork performance indicators, as average routing strategy is compared to current routing strategy. using The models the routing strategies are built such and tested in a travel time, total delay environment time and the and effects of anticipatory routing on suitable differentnetwork vehicle performance types. Evaluation resultssuch show that the microscopic simulation the results are compared using indicators, as average anticipatory routing strategy the network performance, for a limited rate of the vehicles travel time, total delay time can andimprove the effects of anticipatory routingbut ononly different vehicle penetration types. Evaluation results showequipped that the with navigation system basedcan on improve the predictive traffic information, timepenetration compared to reactive reduced anticipatory routing strategy the network performance,the butaverage only fortravel a limited rate of the routing vehiclesisequipped by upnavigation to 20 % insystem the best scenario the vehicles with the predictive navigation system the ones profit routing the mostisfrom the with based on theand predictive traffic information, the average travel time are compared to to reactive reduced predictive strategy. by up to 20routing % in the best scenario and the vehicles with the predictive navigation system are the ones to profit the most from the predictive routing strategy. © 2018 The Authors. Published by Elsevier B.V. © 2019 The Authors. Published by Elsevier Ltd. Peer-review under responsibility of Elsevier the of the mobil.TUM 2018 conference. © 2018 The Authors. Published by B.V. committee This is an open access article under the scientific CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the mobil.TUM 2018 conference. Peer-review under responsibility of the scientific committee of the mobil.TUM18. Keywords: Anticipatory routing; traffic prediction; network performance Keywords: Anticipatory routing; traffic prediction; network performance
* Corresponding author. Tel.: +491736772042. E-mail address:author. [email protected] * Corresponding Tel.: +491736772042. E-mail address: [email protected] 2352-1465 © 2018 The Authors. Published by Elsevier B.V. Peer-review the scientific committee 2352-1465 ©under 2018responsibility The Authors. of Published by Elsevier B.V.of the mobil.TUM 2018 conference. Peer-review under responsibility of the scientific committee of the mobil.TUM 2018 conference.
2352-1465 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the mobil.TUM18. 10.1016/j.trpro.2019.09.082
2
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
495
1. Introduction Urban road networks are experiencing a higher level of traffic and long wasted time in traffic jams every day. Besides the increased level of pollution, higher fuel consumption and large economic losses, traffic congestion causes also mental stress and frustration to the drivers (Eisele et al., 2012). With the rapidly growing population and urbanization, traffic congestions are estimated to worsen in the future. Congestion happens when traffic demand exceeds network capacity. Therefore, a common approach previously used to mitigate congestion was to construct new roads or widen the existing ones, in order to increase the capacity to easily accommodate the increased demand. However, construction of new roads, especially in urban areas, where higher amount of congestion is registered, is firstly costly and it is not sure if it might mitigate congestion, considering the induced traffic that will be caused due to the new constructions. Nowadays three solutions are considered to be able to alleviate congestion: the reduction of travel demand, shifting towards other modes of transport or distributing traffic to better utilize the network capacity (Liang and Wakahara, 2014). As with the increasing population, the number of trips is projected to be increased and not reduced and influencing the drivers to shift to other modes is difficult, using route guidance systems to distribute traffic is seen as the most feasible solution to improve urban traffic jams. The integration of new technologies which provide access to traffic data and the available computational power are major contributors to the successful implementation of route guidance systems, which support congestion mitigation. Intelligent Transportation System attempts to use route guidance systems or adaptive traffic lights to reduce the effects of congestion, by accessing traffic data, via inductive loops, mobile devices or new technologies, such as image processing. Moreover, with the advancement in communication technologies, such as the use of wireless communication, a car has evolved to a sensor platform (Kim et al., 2016). Having greater access to the infrastructure data and more information about the state of the network and possible congestion, vehicle route navigation systems can compute the shortest travel time path based on real-time traffic information, rather than the calculation of the shortest path that was available in the past. Nevertheless, existing route navigation systems provide route guidance based only on real-time traffic information, without considering the future evolvement of traffic in the network. These strategies, known as reactive routing strategies, react only when congestion has already occurred and do to prevent congestion from happening. The future deployment of anticipatory routing strategies, which consider anticipated traffic condition, would probably enhance the effectiveness of the in-car navigation systems, especially in large scale dynamic environments. However, the effects that this routing strategy will have on the network performance, if the route recommendation is provided to a certain number of drivers, are not yet clear. The main goal of this study is to investigate the impacts of anticipatory routing strategy compared to current reactive routing strategy on the network performance, for different penetration rates of vehicles equipped with the predictive navigation device. To compare the results of different routing strategies, the main performance measures that will be used are average travel time and total delay time. Moreover, the impacts of anticipatory routing strategy on different drivers will also be investigated by evaluating the average travel time of different vehicle types. The rest of the paper is organized as follows. Section 2 describes the related work on the topic. Section 3 and Section 4 explain the system model and the simulation setup used to access the results. In Section 5 the evaluation and discussion of the results is performed. Conclusions are described in Section 6. 2. Related Work A lot of research is performed in anticipatory routing and its effects on the network performance. The researchers stress the benefits of anticipatory routing compared to reactive routing mainly in terms of travel time and congestion avoidance, especially when the penetration rate of the vehicles equipped with a navigation system based on the predictive traffic information is high. Following is an overview of the main research works performed on this topic. Kim et al. (2016) in their study compare Dynamic routing with prediction with Static and Dynamic routing. In Static routing the vehicles follow the routes that were assigned to them at the beginning without any change, while in Dynamic routing the route is updated depending on real-time traffic information. Their results indicate that the Dynamic with prediction reduces the travel time effectively compared with the other strategies, especially when the
496
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
3
penetration rate of the vehicles with the predictive navigation device is high. However, after rerouting the first vehicle, they penalize the route that the first vehicle has already chosen, when calculating the route for the second vehicle in the next time step. This is not the case in reality, as it is difficult to take into account the impact of the rerouted vehicle in the network immediately when route guidance is given to the next vehicle. Pan et al. (2013) also studied the benefits of anticipatory routing. They developed five rerouting strategies which proactively compute individually-tailored route guidance to vehicles when there are signs of congestion in the network. Their simulation results show that these strategies can reduce travel time as much as the state of art dynamic traffic assignment (DTA), while requiring less computational time. However, they estimate travel time based on speeddensity relationship using Greenshields’s model, which was initially developed for motorways and does not necessarily hold in urban traffic. To compensate the drawbacks of the modelling approaches which use speed-density relation for traffic prediction in the urban environment, Liang and Wakahara (2014) adopted a microscopic modelling and developed two different prediction methods. The first prediction model was based on spatiotemporal correlation of the actual link and all the other upstream links. This model uses adjustment factors which take into account the effects of traffic lights on traffic dynamics and assumes that the speed is taken from loop detectors rather than derived from conventional speed-density relation. Their second prediction model is based on the approach that the inflow and the outflow on the concerned link are based on the spare load capacity. The results show the benefit of anticipatory routing leading to 70% reduction in average travel time compared to reactive routing. Claes et al. (2011) use a decentralized approach for anticipatory vehicle routing using delegate multi-agent system in which vehicles are routed based on current and predictive traffic information. Their routing approach was inspired by the food foraging mechanism in ant colonies, using pheromones. Multi-agent collect information in the network and distribute it to a central server which computes the best path for a given origin-destination (OD) pair. The experiments show that this strategy helps the drivers who have the predictive traffic information to reach their destination 35 % faster compared to the drivers who do not have data or have real-time traffic data. The approach used in this paper is different from the before mentioned approaches because firstly, it uses a microscopic simulation to estimate the travel time in the network. Moreover, it considers the current procedure of incar navigation systems, when depending on traffic information at a certain point in time, the same route guidance is given to all the vehicles equipped with a navigation device. This technique represents the reality better in terms of vehicle distribution based on a route guidance system, because contrary to the other approaches it does not consider each previously rerouted vehicle one by one in the route recommendation to the next vehicles, which would be the case in an ideal situation. 3. System model 3.1. Basic model A simple test network is built with the help of a microscopic traffic simulation to evaluate the two different routing strategies. DTA is chosen as an assignment procedure as it is more realistic compared to static traffic assignment, allows modelling of variations on travel demand and avoids entering the routes manually (PTV GROUP, 2016; Chiu et al., 2011). Traffic assignment is essentially a model of the route choice in general and the model of this decision is a special case of the discrete choice modelling (Fellendorf and Vortisch, 2010). DTA represents an iterative learning process and calculates the best path for each iteration taking into account the experience from the previous iterations. With the help of DTA, the dynamic user equilibrium can be reached, which means that vehicles departing at a specific point in time among the same OD pair, but choosing different routes, have the same experienced travel time (Chiu et al., 2011). This section describes how the dynamic assignment is performed in this study to achieve a realistic vehicle distribution in the network in the lack of real travel demand data. 3.1.1. Test network and travel time calculation The network tested is represented by nodes and edges. In this representation, intersections are modelled as nodes and roads between the intersections as edges of the network graph. Accordingly, the network area is divided into
4
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
497
subareas and the number of trips between the subareas for a given time interval are defined by a matrix. The travel time is measured for each edge in the network graph. The vehicles report the travel time when leaving the edge. All the travel times reported from the vehicles for an evaluation interval are averaged and the travel time for the edge is calculated. In case of congestion, vehicles also report that they couldn’t leave the edge during that evaluation interval, by reporting their dwell time as well, allowing in this way the detection of heavily congested edges (PTV GROUP, 2016). The travel time of the current iteration is used for path selection for the next iteration considering also the travel time from more distant measurements. The method of exponential smoothing of travel times, with a smoothing factor of 0.5 is used to give 50 % influence to the most recent iteration and less influence to the earlier ones. The smoothed travel time is calculated using the following formula (Fellendorf and Vortisch, 2010): ����� � �� � ������ � � �1 � �� � �������
(1)
Where: � : index of the evaluation interval within the simulation time � : index of the iteration � : index of the edge ������ : measured (observed) travel time on edge � for the interval � in the iteration � ����� : smoothed travel time edge � for the interval � in the iteration � � : the given constant smoothing factor 3.1.2. Path selection
In an abstract graph a path is considered to be a sequence of edges. Therefore, the generalized costs of a path are the sum of the generalized costs of the edges that form that path. In DTA procedure, the best path for an OD trip is found iteratively (Chiu et al., 2011). Firstly, the best path is calculated based on the distance to the destination. In the other iterations, all the available paths to the destination are investigated by giving a zero weight to the paths which were not checked. After all the paths are checked, for each iteration the traffic situation and the travel time on the edges change. Therefore, different best paths are selected until convergence is reached. For the distribution of the travel demand of OD pairs, the Kirchhoff distribution formula is used (Fellendorf and Vortisch, 2010): ���� � � �
���
(2)
∑� ���
Where: �� : the benefit of path � ���� � : the probability that path � is selected � : the sensitivity parameter of the model 3.1.3. Vehicle types
In this study, four different vehicle types are used to represent different navigation systems that the vehicles are equipped with and the usage of these navigation systems. The vehicle types and their general cost functions are described below: 1.
Experienced drivers – normal in-city drivers who do not use navigation system and choose the route only based on experience
��������������1� � �1� � ����������������������
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
498
2.
3. 4.
5
Vehicles with distance navigation system (DIST-NS) – drivers who choose the route that seems to be the shortest depending only on free flow conditions ������� ���� ��� � � � ������ �������� Vehicles with real-time traffic information navigation system (RTTI-NS)
������� ���� ��� � � � ������� ������ ���� Vehicles with predictive traffic information navigation system (PREDTI-NS)
������� ���� �4� � � � ���������� ������ ����
In order to create the basic network setting, initially only the first two types of the vehicles (Experienced drivers and vehicles with DIST-NS) are used. In the next phases, the other two types of vehicles are also considered. 3.2. Reactive routing strategy model This strategy offers route guidance based on real-time traffic information and is the current strategy used to provide route guidance to the vehicles. Fig. 1 describes the procedure used to model this type of routing strategy. After iterative simulations using DTA, the conditions of how the traffic is distributed in reality are assumed to be reached. Then in order to give route guidance to the vehicles with RTTI-NS, static routes is created from DTA. As the network is represented by a node-edge graph, travel time of the edges is needed to be known in order to incorporate it in the algorithm which finds the shortest travel time path.
Fig. 1. Reactive routing strategy model
Travel time for each edge is calculated based on the speed of the edges. During the simulation, the speed for each edge is derived and aggregated in five minutes intervals until the end of the simulation run. The interval of five minutes is chosen because traffic flow and speed oscillate too much on smaller intervals (Guo et al., 2008). Based on the speed and link distance, travel time for each edge in the network is calculated every 5 minutes based on the exact physical relationship:
���� �
� �
Where: t ��� : current travel time d : link distance � : speed on the link
( 3 )
As the network tested is a small network and computation time is not a concern, Dijkstra algorithm (Dijkstra, 1959) is selected as a routing algorithm. Dijkstra algorithm uses the aggregated travel time to find the shortest travel time
6
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
499
path, every five minutes. Then the route guidance is given to the vehicles which are equipped with RTTI-NS. The procedure is repeated for every time interval of five minutes until the end of the simulation. In order to represent the delay in processing traffic information, the shortest route is given to the vehicles five minutes after the shortest travel time path is calculated. For the specific penetration rate of vehicles equipped with RTTI-NS that is selected, a compliance rate of 100% is assumed. 3.3. Anticipatory routing strategy model In anticipatory routing, traffic prediction is integrated into the route guidance system. The predictive travel time is derived using microscopic simulation as a prediction method. The shortest travel time path based on this predictive travel time is then calculated by Dijkstra algorithm and the route guidance is given to the vehicle equipped with the PREDTI-NS. Additionally, route guidance based on current traffic information is provided to the vehicles with RTTINS. 3.3.1. Traffic prediction In order to obtain the predicted travel time for the edges, the parametric prediction method based on microscopic traffic simulation is used. Obtaining the predicted travel time from simulations is meaningful especially in urban environments, due to consideration of the delay caused by the presence of traffic lights and other disturbing traffic. In the case of an incident, as the one tested in this study, using traffic simulation as a prediction tool is assumed to give more accurate traffic predictions results compared to the other prediction methods (Van Lint and Van Hinsbergen, 2012). In this study, predicted travel time data are provided by running the simulation in advance. The simulation that is run in advance, has the same settings as the previously described model of reactive routing strategy. This model, apart from being used as a comparison model, is also used as a basis for the prediction. Fig. 2 illustrates how the simulation of reactive routing model is used as a basis for travel time prediction in anticipatory routing model. During the simulation of reactive routing strategy model, the travel time for each edge is calculated every five minutes, as described in reactive routing section. The traffic information containing the travel time of each edge in the network, every five minutes until the end of the simulation period, are saved. The predicted travel time is derived from the before mentioned saved travel time data five minutes in advance. For instance (Fig. 2), travel time data for each edge in the network, collected during reactive routing model at the point in time ‘ ݐ ͷǯ, is used as predicted travel time data for the edges in the anticipatory routing model at the point in time ‘’ݐ.
Fig. 2. Travel time prediction
Following this procedure, predicted travel time data, showing the state of the travel time five minutes in advance for each edge is available for the anticipatory routing model every five minutes, by using simulation as a prediction tool. 3.3.2. Anticipatory routing model The anticipatory route model considers the predictive traffic information to give the route guidance. As described
500
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
7
in the previous section the predictive travel time is used to calculate the shortest path using Dijkstra routing algorithm and give the guidance to the vehicles. In this routing strategy (Fig. 3), the guidance is given to the vehicles with RTTI-NS and vehicles with PREDTINS simultaneously. The difference is that the vehicles equipped with the predictive navigation system obtain the route guidance based on predictive traffic information and the vehicles with the real-time navigation system obtain the route guidance based on current traffic information.
Fig. 3. Anticipatory routing model
4. Simulation setup The network tested represents an urban grid network, with a surface of around 9 km2 (Fig. 4). The length of the links varies from 200 m to 900 m. All links are two ways links, with one lane per direction. The intersections are all equipped with traffic lights which have a fixed signal program. Traffic demand is generated in the network from four origins and distributed in the network based on four OD pairs. In order to create the initial network setup a realistic distribution of traffic in the network is needed, therefore, as already described, the dynamic traffic assignment is selected as a traffic assignment method.
Fig. 4. Urban network topology in VISSIM
8
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
501
Modeling and testing of the routing strategies are performed in the traffic microsimulation environment VISSIM. Python programing language in combination with COM interface is used to integrate Dijkstra routing algorithm in VISSIM and give the route guidance to the vehicles with RTTI-NS and PREDTI-NS. For a better evaluation of the significance of the results, the anticipatory routing strategy is compared with reactive and fixed path routing strategy. A set of different scenarios considering different penetration rate of different vehicle types are considered for further evaluation. A summary of different scenarios which are tested is shown in Table 1. Scenario F refers to the fixed path routing strategy. In this scenario, the vehicles paths are fixed and do not change when the traffic conditions change. Two vehicle types are present: the experienced drivers and vehicles with DISTNS. Experienced drivers comprise 70 % of the vehicles and choose the route based on experience. Whereas, the vehicles with DIST-NS, which are assumed to take two best routes based on free flow conditions, include 30 % of the vehicles. Scenario R belongs to the reactive routing strategy. Despite the experienced drivers and the vehicles with DIST-NS, the vehicles which have RTTI-NS are also part of the network system in this scenario. These vehicles, which comprise 20 % of the total vehicles, have an updated path every five minutes, based on real-time traffic information. The percentage of experienced drivers and vehicles with DIST-NS is 50 % and 30 %, respectively. Scenario A comprises all the scenarios based on anticipatory routing strategy. In addition to experienced drivers, vehicles with DIST-NS and RTTI-NS, vehicles equipped with PREDTI-NS are also present. The shortest travel time path for this type of vehicles is calculated based on the predictive travel time and their route is updated every five minutes. Scenario A1 until Scenario A6 refer to different penetration rate of this vehicle type, starting from 5 % until 40 % infiltration in the network. The drivers with a real-time navigation and predictive navigation are assumed to comply 100% with the route guidance given from their navigation devices. Table 1. Percentage of different vehicle types for different scenarios. Routing Strategy ___________ Vehicle Type Experienced drivers
Fixed path routing strategy
Reactive routing strategy
70 %
50 %
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
Anticipatory routing strategy
5 % Pred.
10 % Pred.
20 % Pred.
25 % Pred.
30 % Pred.
40 % Pred.
45 %
40 %
30 %
25 %
20 %
20 %
DIST-NS
30 %
30 %
30 %
30 %
30 %
30 %
30 %
20 %
RTTI-NS
0%
20 %
20 %
20 %
20 %
20 %
20 %
20 %
PREDTI-NS
0%
0%
5%
10 %
20 %
25 %
30 %
40 %
Scenarios
F
R
A1
A2
A3
A4
A5
A6
5. Evaluation Based on the indicators of mobility assessment for network performance defined by TRANSPORTATION ASSOCIATION OF CANADA (2006), for the purpose of this study, mean travel time and total delay time, are selected as the main network performance indicators. Moreover, the average travel time of specific vehicle types is also considered in the evaluation. To calculate the average travel time for all vehicles, a weighted average is selected (Madansky, 2017), due to the fact that the average travel time for each evaluation interval is dependent on the number of vehicles for the corresponding interval. The total delay time is obtained by subtracting the actual travel time in a specific time step with the travel time when traveling with the desired speed. An overview of the results for each indicator is presented in this section. The evaluation of the average travel time and the delay time is performed under changing demand conditions, representing a rush hour period and changing capacity conditions, where an incident blocking one lane for 20 minutes is modelled. The travel time was determined for four different OD pairs. The incident that was modelled influenced directly the vehicles following OD 1, partly vehicles following OD 4, and did not directly influence the vehicles following the other two OD pairs.
502
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
9
To capture all these changing conditions the network is simulated for five hours. The evaluation period does not include 30 minutes of warm-up time and 15 minutes of warm-end time. Furthermore, the evaluation is performed separately for the incident period, because the impacts of the rerouting strategies are more obvious during this time frame. The results considered for the analyses were clustered in five minutes intervals. For the purpose of this study, a minimum number of five simulation runs with different random seeds was required for a confidence level of 95 %, based on the procedure described by Dowling et al. (2004). The values from all simulation runs were averaged and were taken into account for further analyses. 5.1. Average travel time Fig. 5 presents the average travel time for all vehicle types during the incident period. During normal free flow conditions, it takes around 12 minutes to perform the trip for each OD pair. However, during the incident period the travel time for reactive routing strategy is increased up to 19 minutes and 15 minutes for OD 1 and OD 4, respectively. These two OD pairs are directly affected by the incident, with more impacts experienced by the vehicles traveling in OD 1. In the best scenario (Scenario A3) the average travel time of all vehicles in the network is around 15 minutes and 13 minutes for OD 1 and OD 4, respectively, showing lower values compared to reactive routing strategy, where the average travel time is 19 and 15 minutes, respectively. Vehicles travelling in OD 2 and OD 3 are not directly influenced by the incident and their travel time during the incident conditions is almost the same as during free flow conditions, therefore their travel time varies slightly compared with reactive routing. For these two OD pairs, the travel time when using anticipatory routing might be even higher than when using reactive routing, because the travel time in free flow conditions cannot be further improved. Additionally, the vehicles which are rerouted interfere with the vehicles coming from other origins and they cause minor congestion in the locations where no congestion was existent before.
Average travel time [min]
Incident period: Average travel time 25 20 15 10 5 0
OD 1
OD 2
OD 3
OD 4
Scenario F
Scenario R
Scenario A1
Scenario A2
Scenario A3
Scenario A4
Scenario A5
Scenario A6
Fig. 5. Incident period: Average travel time
A slight decrease in travel time is noticed for reactive routing strategy (Scenario R) compared to fixed path routing strategy (Scenario F) during the incident period. This stresses the assumption that reactive routing does not help significantly in travel time reduction. As the results of travel time during fixed path and reactive routing strategy are almost the same, only the reactive routing strategy will be considered for further analyses. Fig. 6 depicts the differences in travel time between reactive and anticipatory routing strategy for different penetration rate of vehicles with PREDTI-NS, represented by different scenarios. For vehicles following OD 1, which suffer the effects of congestion the most, anticipatory routing strategy decreases the average travel time of all vehicle types for all scenarios, except Scenario A6. Until 20 % penetration rate of vehicles with the predictive navigation (Scenario A3), the travel time shows a decreasing trend. For Scenario A3, the highest improvement of 20 % reduction
10
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
503
in average travel time compared to travel time during reactive routing, is reached. When the penetration rate is higher than this level (Scenario A4, A5 and A6) the travel time starts increasing again, however until Scenario A5 it is still 10 % less than the travel time for Scenario R. Contrary to the above-mentioned scenarios, when the travel time of all vehicle types decreased compared to reactive routing strategy, in Scenario A6, the travel time of all vehicles in the network is 5 % higher than the travel time during Scenario R. The reason behind this is because all the vehicles that are rerouted take the same route. Consequently, they avoid congestion at the point where the incident happened, but they are also the reason for causing congestion in the routes where they are led to. The reduction in travel time during reactive routing strategy is obvious for all vehicle types. This happens because rerouting of the vehicles with PREDTINS to other routes improves their individual travel time, but also releases the congested links from traffic demand, improving indirectly the travel time of the other vehicle types. For OD 4 reduction in travel time is observed only for Scenario A1, A2 and A3. The highest improvement for this OD is 9 % in travel time reduction and it is reached for Scenarios A1 and A2.
Average travel time comparison [%]
Average travel time comparison with reactive routing (%) 40 30
22
20
5
10 0 ‐10 ‐20 ‐30
‐11‐12 ‐13‐10 ‐20
3 ‐1 ‐5 ‐6 ‐7 ‐2
OD 1 Scenario A1 Scenario A4
33
30
‐2
OD 2
17
11 7 9
‐9 ‐9 OD 3
Scenario A2 Scenario A5
‐1 0
OD 4 Scenario A3 Scenario A6
Fig. 6. Average travel time comparison with reactive routing
5.2. Average travel time of different vehicle types Fig. 7 shows the average travel time for different vehicle types for OD 1, as a better representative, because all the vehicle types considered in the study use this OD pair. Except vehicles with PREDTI-NS, the average travel time of the other vehicle types showed the same tendency during different anticipatory routing scenarios. A decreasing trend was shown until Scenario A3, then the average travel time for experienced drivers, vehicles with DIST-NS and vehicles with RTTI-NS, was increased again reaching even higher values than during reactive routing for Scenario A6. The highest reduction in average travel time for these vehicle types compared to reactive routing strategy was depicted for Scenario A3. For this scenario the travel time for experienced drivers, vehicles with DIST-NS and vehicles with RTTI-NS was denoted to be 15, 17, 13 minutes in absolute values, respectively. Contrary to other vehicle types, the average travel time for vehicles with PREDTI-NS indicated increasing values when the penetration rate of these vehicles was increasing, concluding that the first drivers who will adopt the new technology and will have a navigation device based on predictive traffic information will be the one profiting the most.
504
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
11
Average travel time [min]
Average travel time of different vehicle types (OD 1) 30 20 10 0
Experienced drivers
DIST‐NS
RTTI‐NS
PREDTI‐NS
Scenario F
Scenario R
Scenario A1
Scenario A2
Scenario A3
Scenario A4
Scenario A5
Scenario A6
Fig. 7. Average travel time of different vehicle types (OD 1)
Comparing the travel time of different vehicle types for a specific anticipatory routing scenario, it was noted that for all the considered scenarios, the vehicles which profit the most from anticipatory routing strategy were the vehicles with PREDTI-NS. Their profit was more considerable for lower penetration rates of this vehicles type, with average travel time close to free flow condition, 12 and 13 minutes, respectively for Scenario A1 and A2, which correspond to 20 % and 13 % shorter travel time than for vehicles with RTTI-NS. The vehicles with RTTI-NS are the second ones to profit from anticipatory routing strategy in terms of travel time reduction, followed by experienced drivers and vehicles with DIST-NS. 5.3. Total delay time The total delay time performance indicator is compared with reactive routing strategy and the results are indicated in Fig. 8. It was observed that until a penetration rate of 25 % for vehicles with PREDTI-NS (Scenario A4), the total delay time compared to reactive routing strategy was decreased. The highest reduction in total delay time of 21.9 % is depicted for Scenario A2, when the penetration rate of the vehicles with PREDTI-NS was 10 %. Although the highest profit in travel time for OD 1, which was directly influenced by the incident, was in Scenario A3, for the total delay time, the scenario is different due to consideration of all OD pairs in the network. For Scenario A3 the interference between the vehicles coming from different directions is higher and it influences the lower result in total delay time profit compared to Scenario A2, where the interference is lower due to lower penetration rate of vehicles with PREDTI-NS.
12
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
505
Total delay comparison [%]
Incident period: Total delay comparison with reactive routing strategy [%] 30,0
24,3
20,0 4,9
10,0 0,0 ‐2,3
‐10,0 ‐20,0 ‐30,0
‐10,8
‐11,0 ‐21,9
Scenario A1Scenario A2Scenario A3Scenario A4Scenario A5Scenario A6 Fig. 8. Incident period: Total delay comparison
As expected, considering travel time results, for Scenario A5 and Scenario A6, the total delay time during anticipatory routing is higher than the total delay time during reactive routing. The reason is that the vehicles with PREDTI-NS, when rerouted to other paths trying to avoid the congestion caused by the incident, transfer the congestion in other parts of the network due to high amount of vehicles following the same path. The results are particularly worsen for Scenario A6, when the penetration rate of vehicles with predictive navigation is 40 %, resulting in 24 % higher total delay time than during reactive routing. 6. Conclusion This study evaluates the impacts of anticipatory routing strategy on the network performance with means of microscopic traffic simulation in a grid urban network, in which the traffic prediction is simulation-based. The anticipatory routing strategy is compared to the current reactive routing strategy in terms of average travel time, total delay time and the effects on different vehicle types. The results indicate that the anticipatory routing strategy can improve the network performance until a certain penetration rate of vehicles equipped with PREDTI-NS. Depending on the penetration rate, the improvement can also be significant compared to current reactive routing strategy. In terms of average travel time reduction, the profit is noticed until a level of 30 % vehicles with PREDTI-NS, with the highest profit of 20 % travel time reduction, recorded when the penetration rate of this vehicle type is 20 %. For higher levels than 30 % of vehicles with PREDTI-NS, anticipatory routing strategy does not improve the average travel time to any further extent due to the high number of rerouted vehicles which relocate the congestion to other parts of the network. In terms of total delay time, the advantage of anticipatory routing compared to reactive routing was shown until a penetration rate of 25 % of vehicles with PREDTI-NS, with the highest improvement of 22 % decrease in total delay time when the penetration level of vehicles with PREDTI-NS was 10 %. Vehicles which profit the most from anticipatory routing strategy are the vehicles equipped with PREDTI-NS, because having predictive traffic information helps them to avoid congested links. To sum up, it would be very beneficial from the network performance perspective, to integrate predictive traffic information in navigation systems to give route guidance to the vehicles, but this profit would be possible only until a certain penetration rate of the vehicles equipped with PREDTI-NS. Moreover, other routing strategies are needed to be integrated into the system to support a better distribution of the vehicles in the network and improve the network performance.
506
Aledia Bilali et al. / Transportation Research Procedia 41 (2019) 494–506 Aledia Bilali, Gordon Isaac, Sasan Amini, Nassim Motamedidehkordi / Transportation Research Procedia 00 (2018) 000–000
13
References Chiu, Y.-C., Bottom, Bottom, Jon, Mahut, M., Paz, A., Balakrishna, R., Waller, T. and Hicks, J. (2011), Dynamic Traffic Assignment: A Primer, Transportation Research Circular. Claes, R., Holvoet, T. and Weyns, D. (2011), “A Decentralized Approach for Anticipatory Vehicle Routing Using Delegate Multiagent Systems”, IEEE Transactions on Intelligent Transportation Systems, Vol. 12 No. 2, pp. 364– 373. Dijkstra, E.W. (1959), “A note on two problems in connexion with graphs”, Numerische Mathematik, Vol. 1 No. 1, pp. 269–271. Dowling, R., Skabardonis, A. and Alexiadis, V. (2004), Traffic Analysis Toolbox Volume III: Guidelines for Applying Traffic Microsimulation Modeling Software. Eisele, B., Schrank, D. and Lomax, T. (2012), “TTI’s 2012 URBAN MOBILITY REPORT”. Fellendorf, M. and Vortisch, P. (2010), “Microscopic Traffic Flow Simulator VISSIM”, in Barceló, J. (Ed.), Fundamentals of Traffic Simulation, International Series in Operations Research & Management Science, Vol. 145, Springer New York, New York, NY, pp. 63–93. Guo, J., Williams, B. and Smith, B. (2008), “Data Collection Time Intervals for Stochastic Short-Term Traffic Flow Forecasting”, Transportation Research Record: Journal of the Transportation Research Board, Vol. 2024, pp. 18–26. Kim, K., Kwon, M., Park, J. and Eun, Y. (2016), “Dynamic Vehicular Route Guidance Using Traffic Prediction Information”, Mobile Information Systems, Vol. 2016, pp. 1–11. Liang, Z. and Wakahara, Y. (2014), “Real-time urban traffic amount prediction models for dynamic route guidance systems”, EURASIP Journal on Wireless Communications and Networking, Vol. 2014 No. 1, p. 85. Madansky, A. (2017), Weighted Standard Error and its Impact on Significance Testing, available at: http://www.analyticalgroup.com/download/WEIGHTED_MEAN.pdf (accessed 2 May 2017). Pan, J., Popa, I.S., Zeitouni, K. and Borcea, C. (2013), “Proactive Vehicular Traffic Rerouting for Lower Travel Time”, IEEE Transactions on Vehicular Technology, Vol. 62 No. 8, pp. 3551–3568. PTV GROUP (2016), PTV Vissim 8 User Manual. Transportation Association of Canada (2006), Performance Measures for Road Networks: A Survey of Canadian Use. Van Lint, H. and Van Hinsbergen, C. (2012), Artificial Intelligence Applications to Critical Transportation Issues: Short-Term Traffic and Travel Time Prediction Models, TRANSPORTATION RESEARCH CIRCULAR, E-C168, Transportation Research Board, Washington, D.C.