Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
141 The purpose of this chapter is to identify implementation issues that agencies face as they consider adopting new advanced integrated model tools and approaches. These issues have been grouped into fi ve categories that describe institutional issues, cost and schedule concerns, data requirements, software issues, and application challenges. Follow ing the description of each issue is a set of potential next steps that agencies may consider to address these implementation issues. INSTITUTIONAL AWARENESS AND CAPACITY Knowledge of Tool Capabilities As with any new technology or methodology, a signifi cant challenge to broader adoption is simply that people are unaware of the new capabilities. This challenge is especially true with advanced integrated model systems. Although many of the larg- est MPOs have developed, or are in the process of developing, activity-based model systems, relatively few mid-sized and small MPOs have implemented activity-based models. As a result, one of the reasons for limited knowledge about tool capabilities is that there are relatively few peers that agencies can turn to for guidance and expe- rience with developing and applying these models. This situation is even more acute with regional-scale dynamic network models. Thus far, large-scale dynamic network models have almost exclusively been developed only in research contexts, although a few agencies have recently initiated regional-scale dynamic network model devel- opment efforts. And, at present, no agencies have yet independently developed and applied operational integrated dynamic models. In addition to having a conceptual understanding of the capabilities of integrated dynamic models, it is also important that agency staff have an understanding of how these capabilities can be exploited to support policy and investment analyses. It is also essential that agency staff have an understanding of the limitations and interpretation required in the current applications of existing tools. However, such an understand- 6 IMPLEMENTATION ISSUES
142 Part 2: ISSUES IN ADOPTING INTEGRATED DYNAMIC MODELS SYSTEMS ing is more likely to arise from seeing examples from other regions or from hands-on experience working with integrated models. Thus, there is a conundrum that agencies are less likely to develop and apply advanced integrated models until there are more examples of agencies developing and applying such models. Lack of institutional knowledge may be attributable in part to the nature of how these tools have been used in the past. Activity-based models have primarily been developed as long-range planning tools by MPOs (although some state DOTs have also developed such tools) and are oriented toward providing information on the effects of policy and investment choices such as mode, automobile ownership, work and other location choices, and regional-level static network assignments. In contrast, dynamic network models have more typically been developed on a project-specific basis by state DOTs or by local agencies to evaluate traffic operations and facility design. As a result, agencies may have developed knowledge and experience working with one, but rarely both, of these types of models (AECOM 2010). It is clear, however, that there is significant interest in both the research and prac- tice communities in integrated dynamic model systems, as evidenced by the C10A and C10B projects, as well as independent efforts by agencies such as the SFCTA, MAG, the Chicago Metropolitan Agency for Planning (CMAP), and the San Diego Asso- ciation of Governments (SANDAG). These efforts can inform and support efforts to expand understanding of integrated dynamic model system tools and capabilities. Potential Next Steps ⢠Develop more hands-on experience and sensitivity testing of integrated models, either by building on existing federal efforts such as the SHRP 2 C10A and C10B projects, by building on existing regional integrated dynamic model development efforts led by agencies (such as MAG, SANDAG, CMAP, SFCTA), or by initiating new integrated dynamic model deployment efforts. ⢠Disseminate more knowledge by consultants and agencies about integrated models through outreach efforts such as webinars, conference sessions, peer reviews, and practitioner-oriented documents. ⢠Consult with other agencies that have activity-based models, dynamic network models, or integrated dynamic model systems to better understand model capabili- ties and challenges. ⢠Facilitate information exchange among staff members, within a single agency and across multiple agencies. Staff Resources Developing and applying integrated activity-based models and dynamic networks models requires a broad range of skills. Modelers, analysts, and managers must be knowledgeable in discrete choice theory and other demand modeling techniques, traf- fic flow theory, simulation techniques, computer programming, geographic data man- agement, and statistics, and must also be cognizant of the broad spectrum of concerns, from long-range planning questions to small-scale traffic operations issues.
143 Chapter 6: IMPLEMENTATION ISSUES The overall pool of individuals with relevant experience in developing and apply- ing either activity-based models or dynamic network models is relatively small, and the set of individuals with experience in both domains is even more limited. Agencies that are able to identify and hire staff with the appropriate understanding and skills may still face challenges in ensuring that these staff can be devoted to advanced integrated model development and application and not redirected to fulfill other agency responsi- bilities. This problem may be especially acute in an era of shrinking state, regional, and local transportation planning agency staff. Public agencies also continue to face the challenges of retaining qualified staff because consulting firms often are able to offer more attractive salary, benefits, and growth opportunities to the most highly qualified individuals. Overall, it is challenging for any agency to maintain sufficient numbers of staff members with the necessary skills to competently develop and apply advanced integrated models (AECOM 2010). Ensuring sufficient staff resources is dependent on recognizing the need for the capabilities that integrated dynamic models can provide and necessitates a commit- ment of funding to these efforts. Staff resources can be developed through training and application opportunities, ensuring that knowledgeable staff are retained by agencies, and by establishing means of preserving and transmitting institutional knowledge. Potential Next Steps ⢠Identify dedicated staffing for integrated model development and application efforts. ⢠Strengthen noncompete clauses to disincentivize consultants from attracting skilled modelers from public agencies. Consultant Assistance Public agencies contract with external consultants to provide expertise that they may not have in-house or to supplement the capabilities of agency staff. Such arrangements are advantageous to agencies, because they provide the ability to deploy resources flexibly to where needs are greatest and for limited periods of time. Most agencies have relied on contractors for activity-based model and dynamic network model implementation. This reliance has allowed these advanced models to be implemented more quickly than would have been possible if relying solely on agency staff. In addition, using con tractors facili- tates the transfer of advanced modeling techniques across regions, which is especially critical in emerging areas such as integrated, dynamic models. However, an obvious risk in relying on contractors is that agencies do not develop the necessary staff resources to fully understand, maintain, and apply advanced models. An additional risk is that agen- cies are often reliant on a single consultant rather than having a pool of diverse competi- tive consultants with whom to work. Given the long-term strategic nature of the develop- ment of these models, developing and maintaining institutional knowledge is essential. Overcoming a reliance on consultants can be achieved by ensuring that agency staff are involved in all aspects of model development and application, by providing professional growth opportunities for staff, and by building collaborative relation- ships with other public agencies with whom to share knowledge and experience.
144 Part 2: ISSUES IN ADOPTING INTEGRATED DYNAMIC MODELS SYSTEMS Potential Next Steps ⢠Involve agency staff in all aspects of model development, enhancement, and application. ⢠Include participation of multiple agencies from different regions in parallel inte- grated model implementation efforts. Interagency Coordination Activity-based models and dynamic network models have often been developed by different types of agencies for different purposes. The complexities of developing and applying these models increase further as they are integrated into dynamic model sys- tems, which require substantial technical understanding and significant amounts of diverse data. Many agencies lack either the technical expertise or the data to develop and apply these models, necessitating coordination across multiple agencies. This coordination could involve the sharing of tools, data, or simply experiences, and might involve agencies within a single state or region, or across multiple states or regions. However, achieving such coordination across different agencies can be challenging due to different institutional goals, staff availability, and schedules (AECOM 2010). Despite these differences, agencies can develop strategic partnerships based on exist- ing agency capabilities and on practical common concerns such as cost- effectiveness. For example, primary responsibility for developing and maintaining detailed opera- tional networks may be most effectively led by agencies whose analytic requirements necessitate the use of these detailed data. Potential Next Steps ⢠Define common data standards to facilitate model development. ⢠Include participation of multiple agencies from a single region in a unified inte- grated model implementation effort. ⢠Include participation of multiple agencies from different regions in multiple parallel integrated model implementation efforts. ⢠Develop a forum for agencies to exchange integrated model development experi- ences, data, and tools. COSTS AND SCHEDULE Model Development Costs Costs for developing activity-based models and dynamic network models have been reduced in recent years. On the activity-based model side, these cost reductions can be attributable to a number of factors. First, software development costs have been sub- stantially reduced since the earliest activity-based models were developed as a result of standardization. Model estimation costs have also been reduced as it has become increasingly common for models or model components to be transferred between regions, a practice that that has been validated empirically and statistically. Dynamic
145 Chapter 6: IMPLEMENTATION ISSUES network modeling software has also significantly improved in recent years, with a number of software options providing improved run times, robust user interfaces, and tools to facilitate data development. But despite these software cost reductions and performance improvements, overall development costs for dynamic advanced integrated models are still higher than those for static model systems (AECOM 2010). Some of these higher costs are attributable to data development. Activity-based models require more information than traditional static, trip-based models, such as synthetic populations. Similarly, dynamic network models require detailed network information, such as intersection controls and signal timing. The greatest contributor to these higher costs is the amount of time required to calibrate and test dynamic integrated model components and the overall dynamic net- work model system (Resource Systems Group et al. 2014). In particular, the amount of time required to calibrate and validate regional-scale network components of inte- grated models is significantly higher than the time required for static network models. Because dynamic network models provide a much more detailed and realistic rep- resentation of the transportation network and traffic flows, these models are much more sensitive to small-scale network coding assumptions. As described in the earlier case studies, small-scale coding assumptions can produce global changes in network performance. Iteratively refining network coding assumptions for an entire region can consume significant amounts of agency and consultant resources. Implementing and testing the dynamic integrated model framework is another source of higher development costs. While activity-based and dynamic network assign- ment tools and practices have been improved and refined in recent years, there have been very few efforts that integrate these models. Because methods for integrating activity-based models and dynamic network assignment models are evolving, a signifi- cant amount of resources are required to implement and investigate different integra- tion methods. These investigations must consider issues such as the levels of spatial, temporal, and typological detail that can be exchanged between the model system components, as well as overall model system dynamics, convergence, equilibration, schedule consistency, and other factors directly related to the applicability of these advanced integrated models. While costs for developing both activity-based models and dynamic network models have come down in recent years, these costs may still be prohibitive to some agencies. In addition, costs for developing integrated dynamic model systems are still relatively high given the newness of these approaches. However, the trajectory of reduced costs can be furthered through the development, dissemination, and refine- ment of common software tools and knowledge sharing. Potential Next Steps ⢠Develop, use, and continuously support data preparation and data management tools developed by other agencies. ⢠Perform model calibration and validation automation and documentation. ⢠Adapt existing integration and sensitivity testing strategies.
146 Part 2: ISSUES IN ADOPTING INTEGRATED DYNAMIC MODELS SYSTEMS ⢠Continue research into model transferability. ⢠Include software development training. Model Maintenance Costs All model systems, whether traditional trip-based model systems or advanced inte- grated dynamic model systems, can be expected to require ongoing maintenance (AECOM 2010). This maintenance may involve the updating of networks or socio- economic input assumptions, the recalibration and revalidation of model components to ensure consistency with observed traffic and transit count data, or possibly even the re-estimation of model components to incorporate new travel behavior data. How- ever, dynamic integrated models will likely have higher model maintenance costs for the same reason that they have higher development costs: They require more detailed input data, they are sensitive to these more detailed input data, and the models produce more complex transportation system dynamics that require more time to examine and understand. Even though most MPOs that have implemented activity-based models dedicate staff or consultant resources to the ongoing maintenance of these models, few agencies devote resources to the ongoing maintenance of dynamic network models, usually because such models are developed in support of projects that have defined schedules. Because integrated dynamic model systems are still in the early phases of development, model system maintenance costs are largely unknown, although they can be expected to exceed the combined costs of activity-based models and regional dynamic network models. Potential Next Steps ⢠Use and adapt data preparation and data management tools developed by other agencies. ⢠Ensure there are software maintenance standards. ⢠Consider long-term support contracts with software developers. Development Times Improvements in activity-based model and dynamic network software as well as the availability of detailed data have greatly reduced the amount of time required to implement initial working versions of these models at the regional and corridor levels, respectively. However, it is typical that substantial amounts of additional resources must be invested before each model component, and the overall model system, is ready for application to policy or investment analyses. For activity-based models, lengthy development schedules have typically resulted from the need to collect local travel sur- vey data and other data development, the development of software, the need to cali- brate and validate the new model to acceptable levels, and the need to educate agency staff. For dynamic network models, lengthy development schedules are often the result of the need to iteratively refine network assumptions because of the sensitivity of the networks to small-scale coding changes, the relatively long run times required to reach acceptable levels of equilibration, and the evolution of software to provide new capa-
147 Chapter 6: IMPLEMENTATION ISSUES bilities and handle ever larger networks. In addition to the schedules required to imple- ment the activity-based and dynamic network components, time is also required to design, implement, and evaluate methods for integrating these components. At present, lengthy development times for advanced integrated models may be unavoidable given the emerging nature of these tools and the lack of established prac- tice. However, as the industry moves toward the broader adoption of these tools, development times will be reduced. Model transfers will facilitate quicker model implementations, calibration and validation methods will become better defined, and the research aspects of integrating dynamic model components will diminish. Detailed, user-focused documentation of model development efforts may help facilitate the adoption of these tools. Potential Next Steps ⢠Use model transfers to reduce the need to collect extensive local survey data and the need to estimate local model. ⢠Encourage staff participation and training during model development, so that test- ing and application can occur immediately. ⢠Document advanced integrated model development efforts. Funding Sources In the past, most dynamic network models were oriented toward a specific project or corridor, and the funding for the development of these models was tied to an indi- vidual project. In most cases, dynamic network model development has been funded by state DOTs or by local agencies (AECOM 2010). In contrast, the vast majority of activity-based model development efforts have been regional in scale, and the develop- ment of activity-based models has typically been funded by MPOs as strategic long- term investments that are intended to support a broad range of policy and investment analyses. The different rationales for tool development and the different agencies funding the development of these tools have complicated the development of advanced inte- grated dynamic models. Because agencies have developed activity-based models as longer-term strategic investments, they have been more willing to proceed incremen- tally over longer periods of time, collecting regional household survey data, and mak- ing incremental improvements to data inputs and model components and sensitivities. Funding is typically included as an ongoing budget line item with occasional increases to fund major improvements or data collection efforts. Until recently, most of dynamic network model development efforts have not been designed to support the analysis requirements of multiple projects. While agency experience and institutional knowl- edge increases with each project-specific dynamic network model implementation, the resources devoted to data development, model implementation, and model calibration and validation work performed in support of one project may not directly benefit other projects and may not contribute to the long-term development of a regional- scale dynamic network model.
148 Part 2: ISSUES IN ADOPTING INTEGRATED DYNAMIC MODELS SYSTEMS Agencies like SANDAG and CMAP have recently initiated the development of integrated dynamic model systems as long-term strategic investments, similar to tradi- tional trip-based model systems. These efforts, as well as the case examples of MAGâs Phoenix Inner Loop Travel Model and SFCTAâs DTA Anyway projects, point to shifts in how future model development efforts will be conceived and funded. Potential Next Steps ⢠Consider intra-regional, intra-state, or inter-state funding strategies that support the development of an integrated model framework that can serve local, regional, and multiregion (within state) needs. DATA Data Requirements Activity-based models and dynamic network assignment models provide greater behavioral realism and more detailed information about the impacts of policy and investment choices. In order to provide these enhanced capabilities, advanced inte- grated models require additional types of information beyond those required by tradi- tional trip-based demand models and static assignment models, and this information must be provided at more-fine-grained levels of resolution. Regarding new types of information, dynamic network models incorporate assumptions and parameters that describe traffic flow characteristics such as vehicle effective lengths, saturation flow rates, and response times. Although some of this information can be transferred from other regions or geographical or organizational contexts, in other instances it may be necessary to collect region-specific informa- tion and calibrate these dynamic network model parameters to local conditions. Simi- larly, activity-based models also typically include detailed information not included in traditional 4-step models, such as detailed sociodemographic variables, transit pass- holding, and time-window availability. The resolution of information required for input to advanced models is also greater than required for traditional static models. For example, dynamic network models may require detailed information about intersection controls, signal timing and phas- ing, network loading locations, and street grade. Like static network models, observed vehicle count and speed data are essential, but dynamic network models require that these data be provided for much finer-grained time periods, typically 15 minutes, rather than broad multihour time periods. Although some of these data may be available for subsets of the regional network, developing a comprehensive data set that covers all facility types may be costly and difficult to collect. Also, for both the activity-based and dynamic network components of the inte- grated model system, it is often necessary to synthesize data from multiple existing data sources, such as parcel data, business databases, and existing static model net- works. Whether the data is derived from new data collection efforts, from existing data sources, or from synthesized data, the quality of data is often not known, and
149 Chapter 6: IMPLEMENTATION ISSUES significant time may be spent cleaning existing data or rectifying multiple inconsistent data sources (Resource Systems Group et al. 2014). However, the burden of data development diminishes as integrated dynamic models become more broadly adopted. Data sources and data collection and process- ing tools are becoming better understood and established, and the continued adoption of these methods will only further reduce data requirement burdens. Potential Next Steps ⢠Identify sources of input data and assumptions that must be local and able to be transferred from other regions. ⢠Develop, adapt, and maintain automated data collection, preparation, and syn- thesis procedures or tools. Future and Alternative Assumptions The detailed nature of the inputs to activity-based models and dynamic network models presents challenges when seeking to define future-year or alternative scenario assump- tions in the context of an integrated models system. For example, activity-based model systems may employ spatial inputs defined by fine-grained geography, such as Census blocks and parcels, and fine-grained typological categories, such as employment by detailed industrial sector. Populating alternative scenarios with these levels of spatial or typological detail presents both methodological and institutional challenges. Similarly, dynamic network models use detailed information describing operational attributes such as signal timing and phasing. Observed base-year data describing operation con- trols may not be appropriate for use under conditions of higher future-year demand. This issue is further complicated by the sensitivity of dynamic network components to small-scale coding assumptions. Significant effort is required to debug and optimize different networks. It is critical to ensure that this debug and optimize effort is done in such a way that conclusions drawn from comparisons between alternatives are reason- able (Resource Systems Group et al. 2014). It is inevitable that, as integrated dynamic models are increasingly applied to real-world project analyses, methods for identifying appropriate future and alternative assumptions will become better established. Potential Next Steps ⢠Develop or adapt automated data collection, preparation, and synthesis proce- dures or tools. Data Management and Maintenance Previous sections have described some of the challenges associated with developing the information required to implement and apply integrated dynamic model systems as well as developing these model systems as long-term strategic investments, poten- tially resulting from the collaboration of multiple agencies. As with all models, these model systems require ongoing data maintenance and management. Many dynamic network models have used network inputs that were created on an ad hoc basis for a
150 Part 2: ISSUES IN ADOPTING INTEGRATED DYNAMIC MODELS SYSTEMS particular subarea or corridor. Such development limits the ability of agencies to revise and expand these network inputs as new information becomes available and analysis requirements are expanded. A key issue facing agencies developing advanced models is how to reduce burdens on agency staff while also maintaining and updating the infor- mation that drives the model system. Ideally, such maintenance can be streamlined by the identification or establishment of known data sources and the automation of data collection and data cleaning to the maximum extent possible. Potential Next Steps ⢠Develop or adapt automated data collection, preparation, and synthesis proce- dures or tools. METHODOLOGY AND SOFTWARE Methodology and Software Maturity The case examples described earlier in this part reveal that there are numerous approaches to integrating activity-based travel demand and dynamic network models and that implementing an integrated dynamic model systems involves making myriad design decisions. Agencies and their consultants must consider whether and how to esti- mate local traffic flow models and how to define the spatial, temporal, and typological resolution and segmentation used to link the model system components. Fundamental questions about learning and decision-making processes and notions of convergence and equilibrium must also be considered, and the sensitivities of the model should be systematically evaluated. Tremendous advances in both activity-based travel demand and dynamic network model theory and implementation have been made in the last 15 years. A number of viable software implementations of these models have been realized in the public and private sectors by academic researchers, consultants, and software companies. While this proliferation of software reflects the recognition of the importance of these tools for planning and operation analyses, there are a number of challenges that agencies confront when attempting to select software for inclusion in an integrated dynamic network model system. These challenges include understanding the unique capabilities of the different software options, the quality and performance of these options, and the level of support that can be expected. Agencies need guidance on how to identify the most appropriate activity-based travel demand and dynamic network model components, given their expected appli- cations contexts. For example, most dynamic network models of greatest interest are either mesoscopic or microscopic simulation models. Mesoscopic models consider the movements of vehicles or packets of vehicles using traffic flow theory. Such models often provide faster run times than microscopic models but do not include the same levels of network detail, and thus, may limit the types of alternatives that can be evaluated by an agency (Chiu et al. 2011). In contrast, microscopic models typically simulate the movement of vehicles (and sometimes individuals) using detailed car- following, lane-changing, and gap acceptance models, but usually take longer to run.
151 Chapter 6: IMPLEMENTATION ISSUES A very limited number of network simulation packages offer the ability to model transit pathbuilding and assignment. A key design consideration in developing an integrated dynamic model system is how the primary components will interact and exchange information. This interaction and information exchange involves addressing issues such as the levels of spatial, tem- poral, and typological detail that each model considers, and whether additional modi- fications or capabilities must be developed to facilitate this exchange of information. Because dynamic network model and activity-based model software have been rapidly evolving in recent years, the software code often contains bugs that may not be revealed until alternatives are being evaluated. This potential for software bugs often results in the need to rerun alternatives, which can be problematic given the typically long run times, an issue discussed in a subsequent section. In addition, it may be diffi- cult for agencies to calibrate and validate model components or to identify the sources of problems or unexpected results without the involvement of the software developers. Model features, design, or assumptions are often not well documented. However, because both activity-based model software and dynamic network assignment software have been developing so quickly, agencies can be confident that software capabilities and reliability will continue to improve and will achieve improved levels of maturity. By carefully considering their analytic needs, and by working col- laboratively with software developers, other organizations, and consultants, agencies can help support this process. Potential Next Steps ⢠Assess model usage needs and use of existing documents, agency experiences, and other resources to understand features, capabilities, and performance of existing tools. ⢠Include the participation of multiple agencies from different regions in parallel integrated model implementation efforts, in collaboration with outside consul- tants or vendors. ⢠Provide ongoing software engineering support. ⢠Ensure there is a common, open source, integrated, dynamic model codebase. Computational Resource Requirements One of the most significant challenges to integrated dynamic model implementation is long model run times. Long run times are unsurprising given the tremendous increase in the amount of detail represented in both the demand and supply components of these model systems relative to static model system. However, in practice, these long run times seriously compromise the ability of agencies to use integrated dynamic models for investment and policy analyses. Long run times are the result of a number of factors. The underlying theory guid- ing the structure of each individual model component influences run times, such as whether the activity-based model considers intra-household interactions when simulat- ing daily activity patterns, or whether a mesoscopic or microscopic dynamic network
152 Part 2: ISSUES IN ADOPTING INTEGRATED DYNAMIC MODELS SYSTEMS modeling approach is used. The amount of spatial, temporal, and typological detail also influences model system run times, with increasing levels of detail resulting in longer run times. The level of convergence or stability that is achieved with the dynamic network component, as well as within the overall model system, influences run times, with higher levels of convergence necessary to support detailed project analyses requir- ing more iterations and longer run times. Agencies must carefully consider the types of alternatives that are expected to be evaluated and the detail with which these alterna- tives must be considered as part of the integrated dynamic model design process. Potential Next Steps ⢠Continue ongoing software performance enhancement. ⢠Use cloud computing or establish strategic partnerships with organizations with large computing resources. ⢠Establish automated input data quality assurance/quality control procedures and output data summary and interpretation procedures. APPLICATION Project Requirements Many alternative projects or scenarios could potentially benefit from the enhanced sensitivities and detailed information produced by an advanced integrated model, but a much more limited set of alternatives actually require the use of such an integrated model. However, this limited set of alternatives is quickly expanding as agencies seek new investments intended to manage system performance and demand rather than investments that simply expand physical capacity. Given the significant staff and con- sultant resources and the length of time required to develop an integrated model, agen- cies must carefully consider the trade-offs of the costs of investing in these advanced analytical tools and the benefits of the significantly expanded analytic capabilities. Potential Next Steps ⢠Provide training for an agencyâs decision makers and staff members about the analytic capabilities and limitations of integrated dynamic model systems. Regional-Scale Model Systems As more activity-based travel demand models have been implemented by MPOs, these models are increasingly being used to support investment and policy analyses, and there are numerous examples of such tools being used in conjunction with static net- work models to evaluate regional transportation plan alternatives, congesting pricing alternatives, and other scenarios. Similarly, there are many examples of dynamic net- work models being used to evaluate operational improvements and strategies. How- ever, because so few integrated dynamic model systems have been established, there are fewer examples of how such integrated tools can be applied to evaluate regional
153 Chapter 6: IMPLEMENTATION ISSUES and subregional policy and investment alternatives. Both model systems implemented as part of the SHRP 2 C10 projects were used to evaluate a set of alternative scenarios, but significant research and investigation remains to be done. Potential Next Steps ⢠Make additional integrated dynamic model research, development, and applica- tion efforts. Integration Standards and Procedures SHRP 2 C10 projects focused on integrating two types of tools, activity-based travel demand models and dynamic network models, which have been evolving independently from each other over recent decades. These research efforts demonstrated the feasibil- ity of establishing linkages between these models and created new integrated dynamic network model systems that significantly improve the exchange of data between travel demand and supply models, specifically along the temporal dimension. However, the realm of fully disaggregate and temporally detailed dynamic models is an emerging field with few established rules or conventions. The sequential, iterative processes implemented in the SHRP 2 C10 projects reflects the availability of current tools, but it is not the only possible approach to developing a disaggregate dynamic model. Sig- nificantly more conceptualization, design, implementation, and research is required in order to understand the capabilities and sensitivities of this integration approach and to identify means of improving model system sensitivities. These improvements could range from more comprehensive data exchanges or improved equilibration techniques to entirely new formulations of how people and households structure and revise their travel patterns throughout the day or across multiple days. Potential Next Steps ⢠Document integrated model methods by agencies and consultants. ⢠Research systematic, application-focused evaluations of integrated dynamic net- work model components and integration methods. Calibration and Validation Models require calibration, validation, and sensitivity testing in order to ensure their usefulness as analytic tools. However, advanced integrated model users have reported challenges in calibrating and validating these models and in producing reasonable results. There are a number of reasons for these difficulties. First, the activity-based demand models and dynamic network models that make up the integrated model sys- tem are significantly more complex than traditional static models in the following ways: ⢠They have more components and incorporate more complex linkages among these components. ⢠They operate at fundamentally more detailed spatial, temporal, and/or typological levels.
154 Part 2: ISSUES IN ADOPTING INTEGRATED DYNAMIC MODELS SYSTEMS ⢠They are stochastic simulations. ⢠They often take a long time to run. Second, dome software components include automated calibration and valida- tion tools, but the performance of these capabilities are still being explored. Finally, because these models have only recently been deployed at regional scales, there is a need for established guidelines for the calibration and validation of activity-based models and dynamic network models. Potential Next Steps ⢠Research, test, and document automated model convergence, calibration, and vali- dation procedures, including sensitivity testing. ⢠Research the stochastic aspects of model system performance, including sensitivity testing. Stability Replicability of results is an important quality of any model system. When provided the same input information, model systems should generally produce the same results. This does not imply that the model system must necessarily produce identical results from a single run. In fact, as a result of the stochastic nature of many advanced travel demand and network supply components, a distribution of results from multiple runs provides the most comprehensive indication of model sensitivity. In the integrated dynamic models developed as part of the SHRP 2 C10 projects, the instability of model results was primarily a product of the dynamic network model components. Achieving convergence to stable solutions, if not fully equilibrated solutions, proved challenging, even when the model systems were allowed to repeatedly iterate. This challenge of sufficient convergence can affect the repeatability, stability, and reason- ableness of solutions. However, it must be recognized that stochasticity is intrinsic to the integrated dynamic network models used as well as to the real world. Solution methods that address this variation must be identified and evaluated if these models are to be used in practice. Potential Next Steps ⢠Research and test integrated model methods to ensure sufficient network conver- gence and model system convergence.
