Argonne and its research partners are extending and improving an integrated data analytics platform to solve transportation problems at scale. The approach involves using AI techniques to ingest and condition large-scale transportation-related datasets to improve the underlying travel demand models and the dynamic models embedded in the simulated transportation response model. This will include using neural networks to codify signal dynamics on arterials and inferring the signal phase and timing for arterials with congestive flows from global positioning system (GPS) traces. The project goals are to 

• Improve the previously developed DCRNN model of traffic dynamics for integration into large-scale simulations, using it as a mechanism to remediate congestion impacts. Tasks include developing uncertainty estimates of the predicted traffic flows and speeds, extending the model to ingest large-scale GPS data to allow extension of predictions to larger arterials, and investigating the extent to which transfer learning can allow this model to scale to geolocations in which large-scale data sets are sparse.
• Extend the platform to allow practitioners to evaluate and validate the outcome of scenarios of interest to their city/business. Tasks include going beyond the ML/DL approach that codifies correlation to applying causal inference to provide practitioners with the underlying reasoning behind the outcomes, determining whether the computational models can be transferred from HPC to the cloud while maintaining the computational efficiencies, and developing surrogate models from the large data sets that can be generated using HPC resources.
• Improve existing computational techniques to investigate congestion mitigation solutions. Tasks include using the predictive capability of the DCRNN to detect and respond to events, creating a large-scale control response model that includes communication latency to capture next-generation implementations of traffic control solutions, investigating the impact of reinforcement learning for signal control on large-scale arterials, and including energy impacts of these potential improved control solutions.
• Develop advanced analytics for evaluating city-level impacts of control solutions.

(Image by TierneyMJ / Shutterstock.)