r-Dynamic Simulation-Assignment Model (DynaTAIWAN) Under Mixed Traffic Flows for ITS Applications

ITS ApplicationsTa-Yin HuDepartment of Transportation and Communication Management Science, National Chen Kung University, No.1, Ta-Hsueh Road, Tainan 701, Taiwan, R.O.C.TEL: 886-6-2757575-53224FAX: 886-6-2753882EMAIL: tyhu@.twTsai-Yun LiaoDepartment of Information ManagementChaoyang University of Technology, Taichung, Taiwan, R.O.C.Li-Wen ChenDepartment of Transportation and Communication Management Science, National Chen Kung University, No.1, Ta-Hsueh Road, Tainan 701, Taiwan, R.O.C.Yung-Kuei Huang, Min-Ling ChiangInstitute of Transportation, Ministry of Transportation and Communications, Taiwan, R.O.C.July 31 2006Revised, Nov 15, 2006Number of words: approximately 8,500 wordsSubmitted for presentation at the Annual Meeting of the Transportation Research Board, Washington, D.C., Jan 2007, and publication in Transportation Research RecordITS ApplicationsTa-Yin Hu1*,Tsai-Yun Liao2, Li-Wen Chen1,Yung-Kuei Huang3 Min-Ling Chiang3 1. Department of Transportation and Communication Management Science, National ChenKung University, No.1, Ta-Hsueh Road, Tainan 701, Taiwan, R.O.C.TEL: 886-6-2757575-53224FAX: 886-6-2753882EMAIL: tyhu@.tw2. Department of Information Management Chaoyang University of Technology, Taichung,Taiwan, R.O.C.3. Institute of Transportation, Ministry of Transportation and Communications, Taiwan,R.O.C.ABSTRACTThis research aims at developing an integrated dynamic simulation-assignment model, DynaTAIWAN under mixed traffic flow conditions, for Advanced Traffic Management Systems as well as Advanced Traveler Information Systems. The model is composed of two layers, namely simulation-layer and real-time control layer. The simulation layer is designed to simulate traffic flow patterns according to assumed tripmaker characteristics and/or under a set of given conditions; the real-time control layer receives real-time vehicle information and forecast short-term traffic flow patterns.In this paper, the simulation layer is discussed in detail and numerical experiments are conducted to illustrate functional capabilities of the proposed model. In the simulation process, each vehicle is moved and tracked individually. Four different vehicle types are explicitly considered in DynaTAIWAN, including car, bus, motorcycle, and truck. Vehicles are moving along the link through macroscopic flow relationships, speeds of each type of vehicle are adjusted. Numerical experiments are conducted in a 50-node test network and a Taichung City Network.KEYWORDS: Simulation-Assignment Model, DynaTAIWAN, Mixed Traffic FlowINTRODUCTIONAs a common and important problem in many urban areas around the world, traffic congestion causes tremendous public concern, Intelligent Transportation Systems (ITS) focus on increasing the efficiency of existing surface transportation systems through the use of advanced computers, electronics, and communication technologies. Although several field experiments have been conducted to demonstrate possible benefits of ITS in Taiwan, a dynamic network model needs to be developed to estimate and forecast possible traffic flow patterns in order to efficiently utilize possible traffic control measures as well as advanced traffic information (Hu et al., 2005). Several dynamic models, including DYNASMART and DynaMIT, have been developed to utilize ITS technologies(FHWA, 2001); however, these models cannot be deployed directly in Taiwan because of mixed traffic flow considerations.This research aims at developing an integrated dynamic simulation-assignment model, DynaTAIWAN (Dynamic Traffic Assignment and Information for Wide Area Networks), for Advanced Traffic Management Systems as well as Advanced Traveler Information Systems. The model is composed of simulation-layer and real-time control layer. The simulation layer is designed to simulate traffic flow patterns according to assumed tripmaker characteristics and/or under a set of given conditions; the real-time control layer receives real-time vehicle information and forecast short-term traffic flow patterns.Four different vehicle types, including car, bus, motorcycle, and truck, are explicitly considered in DynaTAIWAN. In order to reasonably reflect these types of vehicles in a road network, two specific characteristics are essential in the model, namely, route choice behavior and vehicular flow behavior. Route choice behavior governs how drivers select their paths and how they respond to different types of travel information, and vehicular flow behavior determines how vehicles are moved in a network with constraints of traffic control strategies. In DynaTAIWAN, vehicle paths in the network are modeled as the explicit outcome of route assignment decisions at origins or at nodes of the network. Traffic flow is represented using a mesoscopic approach where vehicles are tracked individually in mixed traffic flows, and moved consistently with macroscopic traffic flow relations.Since motorcycle is one of the major modes in developing countries, especially motorcycles are heavily used in Taiwan, its impact on traffic flow is critical to the evaluation of traffic conditions in urban areas in Taiwan. In order to reflect the impact of motorcycles, a two-stage method is developed to model motorcycles along the link and at intersections. Motorcycles are moving along the link through macroscopic flow relationships, and the speed of motorcycles is adjusted according to empirical data. At intersections, motorcycles aremoved according to the sequence of left-turn-waiting section, waiting section, and normal traffic lane. The left-turn-waiting area is designed to store left-turn motorcycles for two-stage-left operations. The motorcycle waiting area is located in front of stop line to store motorcycles during the red phase. The movement of motorcycle is discussed in more detail in Section 3. Numerical experiments are conducted in a 50-node test network and a Taichung City Network to explore the dynamic model under different demand levels and real-time information supply strategies.Next section presents a brief review of related models. The proposed framework and associated modeling issues are described in Section 3. Numerical experiments and results from the Taichung City network, used to illustrate important aspects involved in this procedure, are discussed in Section 4, followed by a brief summary.LITERATURE REVIEWIn this section, two important issues are briefly reviewed and described; including the development of TrEPS and basic characteristics of motorcycle.TrEPSTrEPS represents “traffic estimation and prediction systems for real-time applications” and the main purpose is to develop an on-line model to estimate traffic flow patterns under a variety of strategies(FHWA, 2001). The success of ITS deployment depends on the availability of advanced traffic analysis tools to predict network conditions and to analyze network performance in the planning and operational stages. The FHWA R&D of the U.S. initiated a Dynamic Traffic Assignment (DTA) research project to meet these data needs and to address complex traffic control and management issues in the information-based, dynamic ITS environment. Under the DTA Program, two software systems for online traffic prediction and two for offline traffic operations planning were developed and tested. The two online systems, DynaMIT-R and DYNASMART-X, are being field tested at traffic management centers for online traffic management. The offline systems, DynaMIT-P and DYNASMRAT-P, are being used in several region-wide traffic management projects.The details of these two models are briefly reviewed in the following sections.DYNASMARTDYNASMART-P is a tool to support transportation network planning and operations decisions (Mahmassani, 2001; Mahmassani et al., 2003A, 2003B). Two major conceptsdeployed in DYNASMART-P are network assignment models and traffic simulation models. Network assignment models have been applied primarily in conjunction with demand forecasting procedures for strategic (long-term) planning applications and traffic simulation models have been used primarily for traffic operational studies (Mahmassani et al., 2004A).DYNASMART-X, a real-time simulation-assignment model, interacts continuously with multiple sources of real-time information, such as loop detectors, roadside sensors, and vehicle probes, which it integrates with its own model-based representation of the network traffic condition (Mahmassani et al., 2004B). Major functions include (1) Estimate the state of the existing network; (2) Predict future network states, (3) Determine new routing information around problems, and (4) Support Traffic Management Centers (TMC) in the development of real-time route and scenario information for more efficient system management. Consistency checking and updating is an important function incorporated in DYNASMART-X to ensure consistency of the simulation-assignment model results with actual observations and to update the estimated state of the system accordingly. Another external support function is intended to perform the estimation and prediction of the origin destination (OD) trip desires that form the load onto the traffic network and are, as such, an essential input to the simulation-assignment core (Zhou and Mahmassani, 2006).DynaMITDynaMIT-P is the offline version of the real-time traffic estimation and prediction system and can be used for the evaluation of ITS at the planning level as well as for the evaluation of various short-term planning projects (Balakrishna et al., 2005). It models the day-to-day evolution of traffic, traveler behavior and network performance, in response to special events and situations such as incidents, weather emergencies, sports events etc. DynaMIT-P consists of a supply (network performance) simulator, a demand simulator and algorithms that capture the dynamic and stochastic interactions between demand and supply. The supply simulator captures traffic dynamics in terms of evolution and dissipation of queues, spillbacks etc. The demand simulator estimates OD flows that best match current measurements of them in the network and models traveler behavior in terms of route choice, departure time choice and response to information.DynaMIT-R is a real-time system designed to support the operation of Advanced Traveler Information Systems (ATIS) and Advanced Traffic Management Systems (ATMS) at TMC. DynaMIT considers a variety of models of traveler behavior in a detailed network representation structure (MIT, 2002, 2003A, 2003B). Through an effective integration of historical information databases with real-time inputs from field installations (surveillance data and control logic for traffic signals, ramp meters and toll booths), DynaMIT-R (MIT, 2003C) efficiently achieves (1) estimates of network conditions; (2) predictions of network conditionsin response to various traffic control measures and information dissemination strategies, and (3) generation of traveler information to guide drivers towards optimal decisions.Basic Characteristics of MotorcyclesSince one of the important issues in DynaTAWAN is the modeling of motorcycle. Fundamental characteristics of motorcycles are reviewed. Results from field observation made on motorcycle flows in Taiwan are summarized as follows (Hsu et al., 2003):1.Motorcycles are relatively small in size, giving maneuvering flexibility and the freedom tomove and park. Therefore, motorcycles have the agility and the capability to weave through queues in congested areas.2.Motorcycles do not follow the “First In First Out” rule at intersections, and they alwaysattempt to flock to the front of the traffic stream at intersections.3.In Taiwan, motorcycles are prohibited from driving on freeways and expressways.4.Motorcycles are prohibited to make direct left-turns, and they need to follow two-stageleft-turn movements.5.The motorcycle has higher acceleration rate and shorter reaction time than cars at theintersection upon the starting of green time.6.The cruise speed of motorcycle is lower than the speed of car; therefore, the motorcycleplatoon will be overtaken after a few minute departed from the stop line.7.Motorcycles tend to start early at the beginning of green, thus have negative start-up delay.Many motorcycles stop ahead of the stop line and wait for the green phase on the pedestrian crosswalk.8.The saturation flow of motorcycle depends on their queuing behavior near the stop line.FRAMEWORK OF SIMULATION-ASSIGNMENT MODEL: DynaTAIWAN The main purpose of DynaTAIWAN is to simulate and predict traffic flow patterns for mixed traffic flows in order to develop appropriate ATMS/ATIS strategies for traveler information as well as real-time traffic control. Several important simulation-assignment concepts from both DYNASMART and DynaMIT, such as the interactions among vehicles information and traffic systems, are incorporated and extended to consider mixed traffic flows and en-route switching behavior in the development of DynaTAIWAN. Special efforts have been made in modeling mixed traffic flows and enhancing software development processes, including object-oriented design, analysis, and programming to provide flexibility of the system.The overall framework of DynaTAIWAN is shown in Figure 1. The model is composed of two layers, namely simulation-layer and real-time control layer. These layers are described and discussed in details in the following section.Main objectives of DynaTAIWAN are summarized as follows:1.Estimate flow patterns with consideration of individual tripmaker behavior, both pre-tripand en-route switching decisions.2.Estimate flow patterns with consideration of network supply characteristics, includingnetwork geometric configuration and traffic control strategies.3.Predict flow patterns based on real-time flow data through traffic surveillance andmonitoring systems.Simulation LayerThe simulation layer consists of two major components, demand and supply. The demand component processes tripmaker’s decisions on mode, departure time, and route, thus time-dependent OD trip tables are formed. The supply component includes roadway geometric data, traffic network configuration, and traffic control measures. Thus, vehicles and associated paths are generated and simulated in a realistic traffic environment.Vehicles and associated attributes are generated through the demand data, and are loaded into the network according to the selected departure time and routes. Vehicles are moved individually according to prevailing local speeds, consistently with macroscopic flow relations on links. While the vehicles are moving in the network, travel information is provided through Variable Message Signs (VMS) and in-vehicle information systems.FIGURE1. The Conceptual Framework of DynaTAIWANReal-Time Control LayerIn the real-time layer, it is assumed that real-time data from traffic surveillance systems could be obtained in terms of flow and speed, thus the real-time data can be used to generate short-term OD patterns, resulting in OD trip tables. The OD trip tables are loaded into the simulation layer. The predicted flows are obtained through simulation according to the OD tables, thus provide information for advanced traveler information systems as well as real-time traffic control systems. According to the framework, the simulation layer utilizes the real-time flow data as a guidance to adjust simulation processes, thus could predict the flow pattern based on this modification. Since the real-time flow data could not be possible to cover the whole network, flow data from the simulation layer could be used in the absence of real-time data. Two major tasks in the real-time control layer are dynamic origin-destination(OD) estimation and real-time flow prediction. The Kalman filtering technique is used in OD estimation, and the rolling horizon approach is implemented for real-time flow prediction. The real-time control layer will be reported in another paper.Traffic Flow ModelsVehicular movements in the traffic network are simulated through two processes: link movement and intersection transfer.(1) Link MovementMacroscopic traffic stream models are applied in DynaTAIWAN with explicitly consideration of motorcycles. In order to consider the impact of mixed traffic flows, a passenger car equivalent is used to combine all possible vehicle types, such as bus, truck, and motorcycle. The speed is determined macroscopically and assigned to each vehicle; however, speeds of different vehicle types are adjusted according to empirical data. From several field observations, speeds of different vehicle types are varied along the link, for example, truck is much slower than normal passenger car, and motorcycle tends to slower than passenger car. Thus, observation data is used to adjust speeds according to vehicle types.In the link movement process, the Greenshields model, a linear speed-density relationship, is assumed.)1(k k u u jf =(1)Where u represents speed and k represents density; u f is free-flow speed and k j is jam-density. The model is calibrated based on field data for different types of roads, and vehicle are converted to passenger car units. A dynamic model for motorcycle PCE is proposed and described in the next section. The calibrated motorcycle PCE is used to calculate density in mixed traffic flows, and then speeds are obtained for each vehicle type.(2) Intersection TransferThe transfer process performs the link to link or section to section transfer of vehicles at nodes. The transfer process is controlled by several factors: link saturation flow rate, possible capacity, and possible green time. For interrupted link flow, the transfer process appropriately allocates the right of way according to the prevailing control strategy at intersections. Signal control is explicitly modeled, and phases and green times are calculated according to a time clock in the simulation.It determines the number of vehicles that are traversing each intersection in the network at each simulation time step as well as the number of vehicles entering and exiting the network. The results of the transfer include the number of vehicles that remain in queue and the number added to and subtracted from each link for each simulation time step. A wide range of traffic control measures for both intersections and freeways are reflected in the outflow and inflow capacity constraints of the transfer process. DynaTAIWAN provides the ability to explicitly model an array of control elements for bothsurface streets and freeways in urban areas. The major element for surface streets is signal control, which includes pre-timed control and actuated control. Entrance ramp control and VMS are the major controls for the freeway system. Since VMS provide another means to display real-time information, their application is not limited to freeway systems.Multiple Vehicle ClassesFour physical vehicles are explicitly considered in DynaTAIWAN. Two major issues for these vehicles are addressed: how the vehicle moves in the network, and how the vehicle selects the traveled paths. The motorcycle is described in the next section, and the other three vehicles are discussed as follows:1.Passenger CarThe passenger car is traveled according the velocity, V, obtained from the speed-density relationship, and crosses intersections based on prevailing traffic control measures and available capacity. The path selected is based on pre-trip behavioral rules as well as en-route switching rules.2.TruckThe speed of the truck is approximated as V t = t* V, where t is calibrated based on real data. The truck crosses intersections based on prevailing traffic control measures and available capacity, and each truck represents two passenger car units. The path used for the truck can be either the specified route or the route generated based on behavioral rules.3. BusThe speed of the bus is approximated V. The bus is simulated according to specific routes and stops. In terms of capacity calculation, each bus is represented two passenger car units.Modelling for MotorcycleIn DynaTAIWAN, motorcycles are tracked individually. Along the link, each motorcycle is moved according to speeds obtained from the speed-density function. In order to consider motorcycles in mixed traffic flows, several important characteristics need to be identified: estimation of motorcycle passenger car equivalents (PCE), motorcycle moving speed, and motorcycle movement at intersections.1.Estimation of motorcycle PCEAs discussed in the previous section, the linear speed-density function is used in DynaTAIWAN. In order to appropriately capture impacts of motorcycles on traffic streams, motorcycle PCE needs to be determined from field observations. This research extends the//c i i c iV V PCE A A =(2)Where V c and V i = mean speeds for cars and type i vehicles, respectively, in the traffic stream; and A c and A i = their respective projected rectangular areas (length*width) on the road.()()c ii i i d c c d V V PCE W L L W L L =×+×+(3)i d i ii cdc cL L W V PCE L L W V +×=+×(4)Where V c and V i = mean speeds for cars and type i vehicles, respectively, in the traffic stream; and L c and L i = their respective lengths; W i and W c : their respective widths. L d is the length of detection zone, which is the length of observation area (or the length of detectors). The term of effective length divided by speed is equal to occupancy, denoted as O i , which means the total time period occupied by type i vehicles above detection zone.i i i c cW O PCE W O ×=×(5)Where i W and c W are dynamic width for moving vehicles; i O and c O are occupancy for vehicle type i and passenger car. According to Hsu et al. (2003), the physical length and width of a motorcycle are 1.85 meters and 0.75 meters; the physical length and width of a passenger car are 4.00 meters and 1.55 meters. The length and width of a moving motorcycle (50 kph) are 13.30-16.10 meters and 1.14 meters, respectively. The length and width of a moving car (50 kph) are 16.60-19.40 meters and 2.29 meters, respectively.2.Motorcycle moving speedThe speed of motorcycle is denoted as V m = * V . V denotes the macroscopic speed for link and the adjust factor is about 0.8 ~1.0. The ratio is illustrated in Figure 2, in which shows is varying from 0.8 to 1.0 under different vehicle velocities. In order to simplify the simulation process, only a value of is used, though could be expressed as a function of velocity.FIGURE2.Tendency of Velocity Ratio(Motorcycles/Car)3.Motorcycle movement at intersectionsDue to the large amount of motorcycles, several measures are applied to enhance efficiency and safety of motorcycles. Two major approaches to regulate motorcycle movements are (1) design the queue waiting areas for motorcycles in front of stop lines, (2) enforce the two-stage left-turn movement for motorcycles. Since motorcycles form a flock queue ahead of the traffic lanes during the red phase, the queuing waiting area is designed to store such queue, as shown in Figure 3. The motorcycle follows two-stage left turn to avoid conflict between vehicles and motorcycles. As shown in Figure 3,if a motorcycle wants to make a left turn from direction A to direction B, the motorcyclist needs to go straight, stays in a left-turn waiting area, and goes straight when the light turns green for direction B.Through field data, the saturation flow rates for the left-turn waiting area, the queue waiting area, and the normal traffic are about 6, 5, and 3 motorcycles per second.The usage of motorcycle in Taiwan is primarily in the neighbourhood and for shorter trips, thus most of the motorcyclists are familiar with their travel areas. From several field observations, the motorcyclists tend to use the shortest paths (Hsu et al., 2003), so it is assumed that motorcyclists choose the current best paths for their trips. Also, it is assumed that motorcyclists cannot receive real-time traffic information through in-vehicle information systems, though they can still receive information from VMS en-route.FIGURE3. Movements of Motorcycles at IntersectionsPre-Trip Decision ModelsPre-trip descriptive information provides information about the traffic system to tripmakers before they depart from their origins. The first trip decision is mode choice, and possible modes include bus, car, and motorcycle. If the mode of bus is chosen, the tripmaker then determines the departure schedule for a specific bus route. If the mode of motorcycle is selected, the motorcyclist determines the departure time and follows the current best route selection strategy. If passenger car is chosen, the tripmaker needs to determine both departure times and routes. These decisions are made based on the utility maximization theory, associated Multinomial Logit models are calibrated through survey data.En-Route Decision ModelsAs shown in the overall structure, tripmakers follow two different decision models, pre-trip decision and en-route decision models. Pre-trip decisions include mode, departure time, and route decisions; en-route decisions involve possible switching when tripmaker receive real-time information. The en-route decision models are further described in this section.Real-time descriptive information aims to provide traffic conditions to equipped vehicles while they are in the network. En-route decision models reflect possible route switching behavior under real-time information. While en-route, each driver makes switching decisions based several attributes, such as personal characteristics, network familiarity, and information sensitivity.DynaTAIWAN is able to provide 13 different traffic information strategies, as shown inTable 1, via in-vehicle traffic information system and VMS signs. These information strategies are categorized into three types where each strategy provides drivers with different level of detail of the network condition ahead. Thus, the probability of switching routes under the presence of each strategy is different. The default value will be assigned to each of the parameters.In order to reflect possible information supply strategies, three en-route switching models are developed, namely basic model, market segment model, and advanced model. They are summarized as follows:1. Basic ModelThe basic model reflects that drivers can decide whether to make a switch en-route based on received information.=++++=othersU if C d d d d U i i i ,20,1....................*131312122211 (6)Where i U represents the utility of individual i, *i C , represents the choice; 1: switch to a new route; 2: stay in the original route. i d are dummy variables for each supplied information and i are parameters to be calibrated.2. Market segment ModelBased on the basic model, drivers are classified into 3 groups, namely conservative drivers, aggressive driver, and moderate drivers. The same specification across market segments is assumed and the estimation procedure to the different set of data is applied. 3. Advanced ModelThe advanced model considers decision locations, and possible benefits of switching. The decisions locations are classified to near origin point, in the middle of travel, and near destination point, and the locations are defined according to the percent of remaining trip length. The percent of remaining trip length is classified to 0-20%, 20%-80%, and 80%-100% to represent three decision locations. The benefit of switching is defined as saving of travel time. The equation is described as follows:=+++=othersU if C d d d x U i i i ,20,1*44332211 (7)Where i U represents the utility of individual i, *i C , represents the choice; 1: switch to a new route; 2: stay in the original route. 2d ,3d , and ,4d are dummy variables for different。