Hybrid Macro–Meso Simulator¶
Traditionally, the strategic and operational impacts of a scheme or future plan are analyzed by building two separate models: one dealing with planning aspects and the other with operational details.
The planning model is typically a macroscopic static model. It has a simplified consideration of delays and capacities at intersections, theoretical queue propagation, often does not properly capture delays by lane, and contains no evolution of congestion over time.
The operational model is typically a dynamic model, capable of dealing with all these aforementioned aspects but on a smaller scale and often without using equilibrium assignments.
A hybrid macro–meso simulation fuses both techniques in a single model, adopting an innovative approach. The area to be simulated with mesoscopic network loading is defined by a polygon which is then converted to a simulation area. A simulation area can be selected and activated in the Experiment dialog on the Hybrid tab.
The area outside of the polygon is simulated at the same time but with network loading in which travel time depends on a function, like in a macroscopic simulation, instead of depending on behavioral models.
All the centroids in the model produce individual vehicles, which are assigned to a path using either a stochastic route choice path-assignment algorithm or a dynamic user equilibrium (DUE) path-assignment. The calculation of paths addresses the whole model, so all link costs (macro and meso) need to be consistent. How vehicles are assigned in the macro and meso areas, and how costs are calculated are explained in the following sections.
Hybrid Macro–Meso Simulation Process¶
The network model contains macro sections and meso sections. Meso sections are identified by polygons converted to simulation areas; the remaining areas are macro sections. They follow the same rules that are used in the hybrid meso–micro approach: when one input section of a node is inside the meso simulation area, then all input sections are considered as meso. This restriction is needed by the hybrid meso–micro approach and, even though is not strictly necessary for the hybrid macro–meso approach, the same rule is used for consistency across all hybrid models.
There is a unique network that is used to generate individual vehicles throughout the model and it is also used to calculate shortest paths. Shortest paths are calculated using the instantaneous link costs. These link costs use the VDF, TPF, and JDF functions in macro sections and the ICF or DCF functions in meso sections.
The total link cost for all macro links is VDF + TPF + JDF and the total cost for all meso links is either the ICF or the DCF. ICF and DCF are used in the same way they are used in meso, micro, or hybrid simulations. These functions can be selected in the Section dialog and Node (Turn subtab) dialog.
It is your task to define proper cost functions that are consistent for the whole network. The same units, for example minutes, should be used for the output of each cost function, otherwise the calculated paths will not be correct. You can also define a unit-conversion factor in the experiment, which is useful, for example when the VDF, TPF, and JDF functions return a cost in minutes but the ICF and DCF return a cost in seconds.
We strongly advise you to make sure that the functions (VDF or function component) you choose to compute travel time in the macro sections do not produce travel times higher than the route-choice interval. If this does occur, the outflow of a section might drop to zero during an interval, and as vehicles accumulate inside the section, the travel time will increase even further, resulting in progressively more severe blockages. This phenomenon can be triggered, for example, when the capacity of a section is reduced by traffic management actions.
If you do observe these problems, you might need to adjust the functions, the route-choice interval, or both, to find a solution. As a general rule, try to use larger route-choice intervals and seek to avoid exponential growth in travel-time functions.
Before exploring the details of how network loading works in each area of the model, it is important to understand the generation of vehicles at centroids. Whether centroids are in the macro area or the meso area, they all generate individual vehicles following the arrivals model chosen in the experiment.
If the generation centroid is connected to a meso object, the vehicles are generated in the network and immediately start their journey on their currently assigned paths. The volume of vehicles generated depends on the time-dependent demand you have defined, or on the demand estimation process used (see Estimating Travel Demand). The headway between two generated vehicles is determined by the chosen generation model.
If the generation centroid is connected to a macro object, the generation works the same way as described above, the difference being that the vehicles are placed at the boundaries of the first meso area they cross while following the path assigned to them (see Path calculation).
This is done either instantaneously (if instantaneous travel time is selected for the first and last segment in the experiment) or it occurs after the cumulated cost along their path is calculated (if the total VDF/TPF/JDF cost, or a cost component, is selected in the experiment).
The individual generation of vehicles enables:
- Individual tracking of vehicles/trips
- Realistic generation in the meso boundaries
- Consideration of time-dependent generation of traffic following traffic demand profile
- Dynamic assignment over the entire model
- Time-dependent outputs over the entire model
- Application of most traffic management actions in macro areas.
The calculation and selection of paths follows the same rules used by all dynamic simulations. See Dynamic Traffic Assignment Algorithms and DUE for further details. The difference here is that, inside the macro area, instead of using dynamic cost functions, static cost functions will be used (macro VDFs for sections; macro TPFs and macro JDFs for turns).
By default, all Aimsun macro cost functions are expressed in minutes, so in the hybrid macro–meso model you can set a conversion factor to transform the units of macro costs functions to the units used in the dynamic cost functions. As mentioned above, you must define proper cost functions that are consistent for the whole network. You can see the options on the Hybrid tab of the Experiment dialog.
Modeling Hybrid Vehicle Movement¶
Macro network loading¶
In the macro areas of the network, vehicles are treated as individual vehicles rather than using a flow as in traditional macroscopic modeling. The differences compared to meso areas are that individual lanes are not taken into account and travel time is the same for all vehicles going through a section or turn during the same time period. Travel time is evaluated with analytic functions rather than behavioral models.
These functions are volume delay functions (VDFs) for sections and turn penalty functions (TPFs) or junction delay functions (JDFs) for turns. See Cost Functions for more information. If these functions output a generalized cost that includes components beyond the travel time, you should select a function component that provides only the travel time.
As explained in the 'Trip generation' section, vehicles are generated individually at the centroid level. However, to evaluate the travel time for each section, turn, and time interval, the model aggregates the number of trips for each path and assigns flows of vehicles to each macro section and turn.
The value of the flows assigned depends on the changes in traffic demand due to using time-dependent matrices, and also on changes in routing at every route-choice cycle due to congestion in specific paths at a given time interval. This means that the sections and turns of the macro area have time-dependent outputs. There are dynamic flows as well as dynamic costs for each route-choice interval.
Meso network loading¶
In the meso areas of the network, vehicles move according to the usual meso network loading. In summary, the meso area needs to have detailed geometry specifications with correct section lengths, number of lanes, lane connections, a fully detailed control-plan specification, and finally the profiled demand and the specifications for transit. In contrast to the macro areas just described, the travel time is different for each vehicle and is obtained by applying the mesoscopic behavioral models.
When a vehicle is generated in the macro area, it is assigned to all macro sections from its path and it is generated either instantaneously or with a delay in the meso area. The delay on vehicle generation can be specified using either a VDF cost function or a pre-selected cost-function component.
You can define the macro travel time on the Hybrid tab of the experiment dialog. A virtual queue is generated at the boundary where a turn goes from a macro section to a meso section. In this type of turn there is a vertical queue where vehicles coming from the macro area wait to enter the meso area as soon as possible. The delay incurred in the virtual queue is considered in the next meso link. For this reason, these meso entrance sections have virtual queue statistics even if they are not connected to an entrance centroid.
If a non-instantaneous travel time calculation is used, traffic control can be taken into account using the exact method of the Aggregate Control Times (time-specific method). Those control times can be accessed through the DTATurning.
When a vehicle exits the meso area it is assigned to all downstream macro sections. There are two options:
- The vehicle completes its path in the macro area
- The vehicle leaves the macro area and reenters a meso area
In the second case, where the vehicle goes on to reenter a meso area, the vehicle enters through the virtual turn, being the boundary between macro and meso areas. The travel time when moving from meso to meso through a macro area cannot be instantaneous but it is evaluated using the macro functions or function components. You can define the macro travel time on the Hybrid tab in the experiment editor.
You can now use an initial state with hybrid macro–meso models. The following information pertaining to vehicles is stored in the initial state: current section or turn, current lane when a vehicle is in the meso area, current origin and destination centroids, and current path to destination. The initial state also contains information regarding the times when vehicles are expected to exit the network.
When the initial state is loaded, vehicles are placed in the road network in the positions they occupied when the initial state was saved. This means you can load vehicles from meso to macro or from macro to meso areas. The meso areas can be changed, as can the network geometry. The process will load as many vehicles as possible, given the potential changes in the network. Depending on changes, some vehicles might not be introduced.
An important consideration when loading vehicles inside a macro area is that the very first link in which these vehicles move (the one where they were stored) will use the remaining travel time from the stored simulation. This means it does not take into account the current costs of the link. From thereon, the rest of macro-simulated travel will behave as expected.