Data-Driven Energy Use Estimation in Large Scale Transportation Networks

Energy consumption in the transportation sector accounts for 28.8% of the total value among all the industry sectors in the United States, reaching 28.2 quadrillion btu in 2017. Having an accurate evaluation of the vehicle fuel and energy consumption values is a challenging task due to numerous implicit influential factors, such as the variety of powertrain configurations, time-varying traffic and congestion patterns, and emerging new technologies, such as regenerative braking. In this paper, we propose to present a data-driven computational framework to evaluate the energy impact on the transportation system at different scales, leveraging the scalable high-performance transportation simulator, Mobiliti. Instead of using empirical energy models, we create a deep-neural-network mapping between the fuel and energy consumption rate with a variety of heterogeneous driving conditions based on dynamometer test datasets, real-world drive cycle survey datasets as well as real-world GPS probe datasets. For the dynamic driving behaviors, machine learning algorithms are applied over the real-world drive cycle datasets to identify the dominant features and to cluster the drive cycles into representative groups, which can be used to generate high-resolution random drive cycles using a Markov chain approach. Using Mobiliti, both urban-scale static evaluations and dynamic analysis at the trip level can be estimated with a significantly improved fidelity. We demonstrate this approach through case studies with different scales and varied penetrations of different vehicle types, such as conventional ICE vehicles, hybrid vehicles, and electric vehicles.