DYNASMART

DYNASMART (DYnamic Network Assignment-Simulation Model for Advanced Road Telematics) is a discrete time mesoscopic simulation model for ATMIS applications. It is designed to model traffic pattern and evaluate overall network performance under real-time information systems. UCI has also been enhacing the features of DYNASMART for the real-time analysis of advanced transportation managment systems as a core model of the UCI Testbed.

DYNASMART has also been further developed by the University of Texas at Austin, and employed as one of candiadte models for the FHWA's DTA (Dynamic Traffic Assignment) project with DynaMIT. Currently two versions of DYNASMART are avaiable in the DTA project: DYNASMART-X for real-time analysis and DYNASMART-P for planning.

The original capabilities of DYNASMART include:
1) Macroscopic modeling of traffic flow dynamics such as congestion formation and shock wave propagation. Tracking of location of individual drivers.
2) Modeling of different traffic control strategies (freeways, surface streets, signalized intersection, ramp entry/exit etc.)
3) Modeling of prescriptive /compulsory guidance as well as non-prescriptive guidance with trip time information on alternative routes.
4) Modeling of various aspects of the controller such as infrequent updates of network route information database.
5) Modeling of individual drivers' response to information in the case of descriptive guidance based on a set of paths rather than a single shortest path. Random assignment of driver behavioral characteristics. Flexibility to incorporate alternative behavioral rules.
6) Modeling of capacity-reducing incidents at any time, anywhere in the network.
7) Modeling of cases with only a fraction of the vehicles equipped for information.
8) Capacity to carry out simulations based on externally specified dynamic equilibrium paths for drivers not equipped to receive information.
9) Several levels of output statistics for the system, for individual drivers as well as for groups of drivers (equipped drivers, unequipped drivers, drivers on certain O-D pairs etc.). Statistics include average trip time, distances, average speeds and a variety of routes switching statistics.

The basic DYNASMART simulates individual vehicles along the paths selected out of k-shorest paths, and vehicles make its route decision based on their characteristics and information acquisition capability. The following diagram shows how vehicles are moving in DYNASMART.

DYNASMART for Dynamic Optimal Route Guidance and Control

Various information/route guidance systems are modeled in DYNASMART-UCI with following considerations.
· Types of Information Devices: IVNS, CMS, Radio information
· Use of Information: Market penetration, compliance rate (or level of reliability)
· Types of Routing Strategies: Instantaneous, Predictive (UE/ SO)
· Information accuracy: Data aggregation and information update interval

Four types of optimal route guidance are modeled reflecting different optimal route selection procedures:
· TYPE I: Instantaneous information
· TYPE II: Predictive information with information discounting strategy
· TYPE III: Dynamic user optimal routing (average cost routing)
· TYPE IV: Dynamic system optimal routing (marginal cost routing)

These four types do indeed capture most of the existing and envisaged IVNS in the world. Type I and II are straightforward in their modeling. In the Type I, the shortest route is selected directly from the uses most updated travel time information. Type II seeks the time-dependent shortest path using projected time-dependent travel times. For the time-dependent travel time projection, the information discounting strategy is employed. Route guidance schemes of Type III and IV are obtained via multi-user class dynamic traffic assignment procedure. Route guidance schemes are represented by number of vehicles departing from origin i to destination j during time period t via path k. Difference between Type III (dynamic user equilibrium, DUE) and Type IV (dynamic system optimum, DSO) is the objectives applied. While the resulting traffic pattern from Type III minimizes individual vehicles travel time, that from Type IV minimize the total travel time spent in the network by all vehicles. The route guidance schemes for Type III and IV are obtained via iterative approach. Every iteration optimal paths are found via time-dependent shortest path routine. Time-dependent average travel times are used for DUE, whereas time-dependent marginal travel times are used for DSO. The solutions for this type of route guidance are based on the full knowledge assumption.

Real-time Control Systems in DYNASMART

Control systems in DYNASMART can be classified into several classes in terms of their algorithmic structures. Their algorithmic structures are decided by adopting one type from three categories:

• user optimal vs. system optimal
• instantaneous vs. predictive
  
• open-loop vs. closed-loop system

When the second and third categories are considered, four routing approaches exist as follows.

• Simple feedback approach: instantaneous travel time based feedback system
  
• Simple predictive approach: historic data based prediction system
  
• DTA approach: dynamic traffic assignment based open-loop system
  
• DTA feedback approach: traffic assignment based feedback system

Simple Feedback Approach

The simple feedback approach seeks optimal routes based on instantaneous travel time at every instant. The control interval is dependent on the capacity and speed of data processing. The main drawback of the system is that the "optimal" routes at various instances may not be really optimal when reviewed after trip. However, this approach does not require any other input other than link travel times. Measurement errors in network monitoring system are the only source of disturbance. Other disturbances including drivers' compliance rate are reflected through network conditions. This approach can also be applied independently without conflict with other infomation devices. That is, advantage of this approach is not only that it is simple and easy to implement but that it is also robust in application.

Simple Predictive Approach

With increasing interest in dynamic route guidance, there have been many studies on short-term travel time prediction. Travel time prediction models studied include time series analysis, Kalman filtering models, neural network models, and historical and real-time profiles .The simple predictive approach utilizes historical traffic data and real-time instantaneous traffic data. The historic information is insufficient for drivers to find the optimal routes due to day-to-day traffic variations. This method is based on a reasonable assumption that the traffic information on the very next link in the network is reliable but the current information on the farther areas will be less useful due to the temporal variation of the link travel time.This approach reflects temporal variations in traffic condition; however, it does not take drivers' route changes due to the information. Success of this approach depends on the accuracy of historical travel time data and similarity of daily traffic pattern.

Predictive DTA approach

The dynamic traffic assignment (DTA) approach is to optimize the network under the assumption of full knowledge. Unlike instantaneous feedback approach, this approach provides user optimal paths which are expected to be user-optimal after the drivers "experience" the travel not just at an instant. In this approach there is no feedback routine, and so this approach is classified as an open-loop system that relies on perfect prediction. This approach assumes that all prediction components are perfect. In fact, any imperfection of the approach may worsen the system. In reality, there exist errors in every component, such as O/D demand prediction, the route choice model, the compliance prediction, and the network flow model. Therefore, this approach is hard to apply in real-life unless the model is well validated, despite its attractiveness from a theoretical point of view. This predictive approach can be used for both user optimal and system optimal by applying average cost for user optimal and marginal cost for system optimal. Similarly the proposed heuristic can be used for both cases. In the dynamic case, it is almost impossible to find the true marginal costs due to time-dependency, so that quasi-marginal costs are calculated and used for marginal cost routing.

DTA Feedback Approach

The DTA feedback approach is similar to the rolling horizon version of predictive approach in the sense that it seeks an optimal solution for demand during a time interval via iterative method. However, this predictive feedback approach optimizes parameters of each model components. This approach brings a lot of issues with respect to on-line model calibration and interaction between model components. The main issue of the approach is how to reflect stochastic nature of the system into the prediction model. Some model parameters may need on-line adjustment while some of them can be adjusted later. Difficulty of the adjustments arises from the fact that the observed errors are a mixture of each model components' errors. Basically, the aim of the approach is to minimize the gap between the simulation model and real world.