Bae, Junseo (2017-05). A Multi-Contextual Approach to Modeling the Impact of Critical Highway Work Zones in Large Urban Corridors. Doctoral Dissertation. Thesis uri icon

abstract

  • Accurate Construction Work Zone (CWZ) impact assessments of unprecedented travel inconvenience to the general public are required for all federally-funded highway infrastructure improvement projects. These assessments are critical, but they are also very difficult to perform. Most existing prediction approaches are project-specific, shortterm, and univariate, thus incapable of benchmarking the potential traffic impact of CWZs for highway construction projects. This study fills these gaps by creating a big-data-based decision-support framework and testing if it can reliably predict the potential impact of a CWZ under arbitrary lane closure scenarios. This study proposes a big-data-based decision-support analytical framework, "Multi-contextual learning for the Impact of Critical Urban highway work Zones" (MICUZ). MICUZ is unique as it models the impact of CWZ operations through a multi-contextual quantitative method utilizing sensored big transportation data. MICUZ was developed through a three-phase modeling process. First, robustness of the collected sensored data was examined through a Wheeler's repeatability and reproducibility analysis, for the purpose of verifying the homogeneity of the variability of traffic flow data. The analysis results led to a notable conclusion that the proposed framework is feasible due to the relative simplicity and periodicity of highway traffic profiles. Second, a machine-learning algorithm using a Feedforward Neural Networks (FNN) technique was applied to model the multi-contextual aspects of iii long-term traffic flow predictions. The validation study showed that the proposed multi-contextual FNN yields an accurate prediction rate of traffic flow rates and truck percentages. Third, employing these predicted traffic parameters, a curve-fitting modeling technique was implemented to quantify the impact of what-if lane closures on the overall traffic flow. The robustness of the proposed curve-fitting models was then scientifically verified and validated by measuring forecast accuracy. The results of this study convey the fact that MICUZ would recognize how stereotypical regional traffic patterns react to existing CWZs and lane closure tactics, and quantify the probable but reliable travel time delays at CWZs in heavily trafficked urban cores. The proposed framework provides a rigorous theoretical basis for comparatively analyzing what-if construction scenarios, enabling engineers and planners to choose the most efficient transportation management plans much more quickly and accurately.
  • Accurate Construction Work Zone (CWZ) impact assessments of unprecedented travel inconvenience to the general public are required for all federally-funded highway infrastructure improvement projects. These assessments are critical, but they are also very difficult to perform. Most existing prediction approaches are project-specific, shortterm, and univariate, thus incapable of benchmarking the potential traffic impact of CWZs for highway construction projects.

    This study fills these gaps by creating a big-data-based decision-support framework and testing if it can reliably predict the potential impact of a CWZ under arbitrary lane closure scenarios. This study proposes a big-data-based decision-support analytical framework, "Multi-contextual learning for the Impact of Critical Urban highway work Zones" (MICUZ). MICUZ is unique as it models the impact of CWZ operations through a multi-contextual quantitative method utilizing sensored big transportation data.

    MICUZ was developed through a three-phase modeling process. First, robustness of the collected sensored data was examined through a Wheeler's repeatability and reproducibility analysis, for the purpose of verifying the homogeneity of the variability of traffic flow data. The analysis results led to a notable conclusion that the proposed framework is feasible due to the relative simplicity and periodicity of highway traffic profiles. Second, a machine-learning algorithm using a Feedforward Neural Networks (FNN) technique was applied to model the multi-contextual aspects of iii long-term traffic flow predictions. The validation study showed that the proposed multi-contextual FNN yields an accurate prediction rate of traffic flow rates and truck percentages. Third, employing these predicted traffic parameters, a curve-fitting modeling technique was implemented to quantify the impact of what-if lane closures on the overall traffic flow. The robustness of the proposed curve-fitting models was then scientifically verified and validated by measuring forecast accuracy.

    The results of this study convey the fact that MICUZ would recognize how stereotypical regional traffic patterns react to existing CWZs and lane closure tactics, and quantify the probable but reliable travel time delays at CWZs in heavily trafficked urban cores. The proposed framework provides a rigorous theoretical basis for comparatively analyzing what-if construction scenarios, enabling engineers and planners to choose the most efficient transportation management plans much more quickly and accurately.

publication date

  • May 2017