Conference Papers 2022

Seminar
Day 2 (8 Sep 2022), Session 4, ACTIVITY BASED MODELS, 10:00 - 11:30

Status
Accepted, awaiting documents

Submitted by / Abstract owner
Peter Vovsha

Authors
Peter Vovsha, Bentley Systems
Gaurav Vyas, Bentley Systems

Short abstract
To what extend could an ABM add something to the evaluation of transportation policies on top of what can be done with a simpler 4-step model? The paper provides new evidence on the differences between 4-step models and ABMs in a systematic way.

Abstract
Current state of the practice in travel modeling is characterized by a co-existence of aggregate trip-based 4-step models (in practice for over 50 years around the world) and a newer generation of Agent-Based Models (ABMs) implemented in a disaggregate microsimulation fashion that opens a way to utilize tour as the main unit of modeling (in practice for over 20 years, especially common for major cities in the US). While the conceptual advantages of ABMs are well known and generally accepted by the modeling community, they were frequently formulated in such terms as “behavioral realism” or “internal consistency” that are less tangible for a wider community of transportation planners. A legitimate question is to what extend an ABM could add something to the evaluation of transportation projects and policies on top of what can be done with a simpler 4-step model. It is welcome to demonstrate the differences between two models on a series of realistic project examples for the same region and based on the same main inputs.
Until now direct comparisons were very limited and rarely reported in any convincing way. For most model developers and users, the choice of the model type was largely “either-or”. Development and calibration of a complete regional 4-step model, not talking about ABM, requires a substantial effort and budget. Agencies that decided until now to stay on the 4-step model obviously cannot contribute to this comparison. Agencies that switched to an ABM (more than 20 major cities in the US) had abandoned their 4-step models in most cases. Thus, they can witness the viability and practicality of the ABM but cannot really prove that it has much of an added value over a simpler 4-step model that could have been used instead. The luxury of having two types of travel models in a “good shape” and ready for comparisons with the same inputs, until now was beyond the reach for either practitioners or researchers.
Even if the two models are available it is a non-trivial task to make an objective comparison between them. The comparison should be focused on the pros and cons of the model structure rather than on how well the models were calibrated. This calls for a substantial effort in validation and calibration of each of them. However, this effort should avoid any artificial overspecification or pivoting that would mask the quality of the model. Another difficulty is associated with different levels of segmentation pertinent to these models. 4-step models are inherently limited in terms of number of trip purposes and population segments while ABMs can accommodate any segmentation. Thus, apples-to-apples principle is almost impossible to apply.
The current research addresses these issues and provides some new evidence on the differences between 4-step models and ABMs in a systematic way. The research utilizes the recently developed software Emme Agent that is a flexible platform for an efficient assembly of different types of travel models. It includes typical prototype structures for a 4-step model and ABM that can be run with the same socio-economic, land-use, and network scenario inputs that facilitates a side-by-side comparison of the results. The model segmentation (such as trip purposes and modes) was held equal between the two models to the extent possible. However, in some cases, especially when it comes to temporal resolution, the gap between two models was unavoidable and should be recognized as an integral feature of the model structure.
Both models were calibrated to a very good degree of replication the target traffic counts and transit ridership for the Base scenario for the city of Winnipeg, MA chosen as an example. The test projects for model sensitivity were chosen amongst the most typical ones with an addition of some new modeling aspects associated with new mobility. They included a new rail transit line with a differential frequency by time of day, road pricing differentiated by time of day, and mixed land use scenario. For each test, the 4-step model response was compared to ABM response with the analysis and explanation of the differences.
The following main conclusions were made. When a simple infrastructure extension or entire-day policy such as flat tolls is applied, two models most frequently produce similar results. However, when the policies become more elaborate in terms of time-of-day differentiation, the differences between two models become prominent and, in general, the limitations of 4-step model become transparent. The authors believe that it is important to recognize these specifics of model types and systematic biases of the 4-step models compared to ABMs.

Programme committee
Transport Models