Route Assignment in Paramics Estimator
From Traffic Analysis and Microsimulation
Matrix Estimation (ME) is a set of techniques applied by advanced microsimulation users to create synthetic Origin-Destination (OD) matrices. Matrix Estimation is a fallback or second-best option when a local OD study is unavailable or infeasible due to project time or budget constraints. Matrix Estimation is a mathematical problem with multiple solutions, and as a result creating a realistic synthetic matrix for a network requires both art and science.
About the Tool
Paramics Estimator is a software product that is sold separately from the rest of the Quadstone Paramics suite. Estimator creates synthetic OD matrices using traffic counts and a pattern matrix supplied by the user.
The pattern OD matrix (also called a prior matrix or a priori matrix) strongly influences the "shape" of the final matrix, so it is important to have a realistic pattern as a starting point. It is also important to give Estimator as much traffic count data as possible. Intersection turning count movement counts and the mainline counts at the zone boundaries are particularly influential in the ME process. Any location where Estimator is not given a target volume is "uncontrolled." At uncontrolled links the software has no boundaries on what volume is reasonable, so the person who is running Estimator must review each uncontrolled link manually to make sure the volume that has been assigned to it is plausible.
During matrix estimation, Estimator uses the same network that is defined in the primary modelling software (Paramics Modeller) including any model objects that influence vehicle movement, lane usage, flow control and priorities, and route choice. Like the rest of the Paramics suite, Estimator has two routing logic choices: ALL OR NOTHING or DYNAMIC ASSIGNMENT. In addition, varying degrees of PERTURBATION may be applied. In effect this provides four different route assignment combinations:
- ALL OR NOTHING without PERTURBATION
- ALL OR NOTHING with PERTURBATION
- DYNAMIC ASSIGNMENT without PERTURBATION
- DYNAMIC ASSIGNMENT with PERTURBATION
Routing is also affeted by the GENERALIZED COST COEFFICIENTS that have been set in the CORE NETWORK ATRIBUTES --> CONFIGURATION --> ADVANCED tab. By default, vehicles choose their path based 100% on the route which has the shortest travel time. This setting can (and should) be adjusted to take into consideration the travel distance. In addition, a factor for tolls should be included if any toll facilities exist in the network. These settings have a strong influence on vehicle routing in networks where there is a lot of route choice. Please see the Suggested Paramics Settings for additional guidance.
ALL OR NOTHING Assignment
ALL OR NOTHING (AON) assignment is similar to Wardrop Equilibrium. In this method, all vehicles operating between a given O/D pair are routed along the "best" (shortest, fastest, or cheapest) path, and no other paths are ever used by vehicles assiged to that O/D pair. The relative weighting of distance, time, and cost can be adjusted by the user. If PERTURBATION is enabled, a portion of the vehicles will deviate from the best path, mimicking the effects of drivers who take slightly longer routes due to variations in knowledge/perception of travel conditions and personal preferences.
If the Quadstone Paramics network being developed is using ALL OR NOTHING assignment, the network does not need to be perfect in terms of lane usage, signal timings, and JUNCTION PRIORITIES (unsignalized intersection traffic control) as long as the primary routes used between each O/D pair are shown to be correct. This feature can help expedite model development.
When FEEDBACK is enabled, Quadstone Paramics uses Dynamic Assignment techniques: vehicles classified as "familiar" periodically to re-evaluate their route choice based on current traffic conditions, while those classified as "unfamiliar" stick to the "best" path. Inititally, all vehicles are routed along the "best" path (similar to the ALL OR NOTHING technique), but after a time some of the familiar vehicles in the network will reconsider their routes based on current information (such as travel time). Presumably, this is similar to the behavior of real-world drivers that receive real-time traffic information from radio traffic reports or in-vehicle GPS.
If Estimator is used for a network with DYNAMIC ASSIGNMENT, the network must be very well coded and audited prior to matrix estimation. Any incorrect network coding will strongly influence the O/D matrix estimation process, possibly resulting in matrix errors or failure to converge to a stable matrix. For example, if the modeled lane allocation or signal timing at an intersection is incorrect and results in congestion at the interection, vehicles might re-route to avoid the intersection in an unrealistic way. In the worst case, traffic assignment will oscillate wildly between FEEDBACK cycles, resulting in surges of congestion followed by several minutes when the volume on the same route is extremely light. To avoid such problems, the O/D matrix should initially be developed with ALL OR NOTHING assignment and FEEDBACK should not be enabled until a basic, stable matrix has been obtained.
Routing Differences Between Paramics Estimator and Other Matrix Estimation Tools
Traditional matrix estimation techniques that use path based volume averaging to achieve a convergence criteria may not be able to operate with an imperfect source network for ALL OR NOTHING assignment. During estimation, most traditional techniques rely on link/intersection capacity estimates that are affected by intersection traffic control and signal timings.