Exploratory modeling with travel demand models requires capabilities to run models many times, and to process and visualize the results. A full-scale travel demand model may take hours or days to complete a single model run. Using TMIP’s Exploratory Modeling and Analysis Tools (TMIP-EMAT), fast-running metamodel versions can be created that closely approximate the results from the larger models. An alternative approach is to simply use much faster running strategic planning travel models, which forsake explicit network modeling in favor of simplicity and speed. VisionEval (VE) is an increasingly popular framework for this, enabling a number of related models to share a common set of formats and tools. The various models implemented in this framework (e.g. RPAT, RSPM) run moderately quickly, taking minutes instead of hours to finish.
Both VE and TMIP-EMAT include capabilities to manage model runs for multiple scenarios. The VE approach includes swapping out various model input files with alternate modified versions. This allows VE to describe various categorical “levels” for inputs. For a limited number of dimensions, this approach is both effective and easy for to understand. However, as the number of input dimensions and the number levels in each dimension increases, the number of model runs needed expands exponentially. By contrast, TMIP-EMAT allows (indeed, prefers) continuous variation of inputs, and includes methods to efficiently explores that continuous input space, with a focus on metamodel support.
To leverage the particular strengths of each of the frameworks, we developed a TMIP-EMAT interface module for VE’s RSPM model based on the Rogue Valley in Oregon. The implementation expressed 13 unique input dimensions, including variation in the value of time, income levels, land use, transit service, gas and electric costs, and more. By using TMIP-EMAT’s automation tools, we ran a set of 130 experiments designed to create a metamodel for the RSPM. Cross-validation scores for ten selected output performance measures (including greenhouse gas emissions, transit ridership, VMT, and more) exceeded 0.99 across all dimensions, indicating an extremely good fit between the RSPM model and its metamodel counterpart. Interactive exploratory tools were connected to the meta-model, enabling comparisons of risk between scenarios.
This work joins together two important exploratory modeling tools, leveraging the strengths of each. The VE platform offers a moderate level of detail and an interface that is broadly portable between regions, such that adapting the TMIP-EMAT interface for use with a different VE model will be relatively trivial. The metamodel version of RSPM evaluates fast enough to be used in dynamic exploratory tools, enabling the use of a broader spectrum of exploratory modeling methods, including scenario discovery and robust optimization methods.