Machine Learning for Individualized Vehicle Eco-Routing

Vehicle “eco-routing” means that instead of driving the fastest or shortest route between two locations, a driver would choose the most energy-efficient route. Eco-routing operates at the macroscopic level to identify the most energy efficient route to a destination based on the specific vehicle and powertrain. Using coarse information about traffic congestion, grade, stop signs, school zones, charging opportunities, etc., it may be found that the shortest or fastest route is not the most energy efficient. Recent prior work found instances of 20% energy savings using eco-routing. In many cases, the travel time penalty was minimal. The most impressive eco-routing energy savings are based on a highly accurate vehicle and powertrain model. In order to generalize eco-routing technology for widespread use to include different vehicles and powertrains, it becomes extremely important to validate prediction models. Development of highly-accurate a priori powertrain models for every possible vehicle over every possible drive cycle is financially and logistically impossible. Therefore, the research team proposes to utilize Machine Learning (ML) techniques to synthesize individualized models of energy consumption using on-vehicle data collected during real-world driving experience. The overall goal of Phase I is to demonstrate the feasibility of using ML techniques to enable effective, energy-saving eco-routing without a priori detailed vehicle and powertrain models. This technology will enable the rapid, widespread adoption of eco-routing technology and facilitate significant energy savings with commensurate financial and environmental benefits. In Phase I, the team will train and evaluate ML-based eco-routing algorithms using a combination of real-world driving data collected on previous projects and simulated driving data using high-fidelity models. In Phase II, the team will further refine the ML-based eco- routing algorithms as well as develop prototype software products to set the stage for Phase III commercialization. The technology will be marketed as a software development kit (SDK) for integration into third-party navigation applications/systems and also as a metered, pay-per-use web service. Effective, widespread eco-routing technology adoption could potentially save 10% or more of energy consumed by on-road vehicles with corresponding economic and environmental benefits.