SBIR Phase I: Automated Learning of Vehicle Energy Performance Models

点击翻译按钮获取中文摘要

This Small Business Innovation Research (SBIR) Phase I project will research an internet-of-things (IoT) platform to automatically learn vehicle energy performance models (VEPMs). VEPMs are used to predict driving range and battery state of health in electric vehicles (EVs) on a per-vehicle, per-driver, per-route basis, and 8-10 times more accurately than today. It is estimated that over $150 billion will be invested in the electric vehicle ecosystem over the next decade. A significant obstacle hindering rapid EV adoption is range anxiety representing user concerns over the achievable distance and where/ when to charge the EV. Range anxiety can be alleviated by providing EV drivers with contextual intelligence on their realistic driving range and recommended charging strategy, based on travel plans, driving behavior and vehicle model. Increasing EV adoption by consumers reduces transportation system fossil fuel consumption and emissions. As part of this effort, a cloud application programming interface (API) will deliver predictions based on the learned VEPMs; this will also enable energy-aware applications such as eco-routing, eco-cruising, eco-powertrain control, and planning of charging stops, among others, both at the individual vehicle and at the fleet level. Energy-aware applications can increase the overall energy efficiency of electrified fleets.The intellectual merit of this project is to advance an IoT architecture to automatically learn VEPMs from real-time vehicle sensor telemetry and other data, such as maps and route topography. The plan is divided into three integrated goals: (1) the building of an IoT framework leveraging physics principles to capture the vehicle motion and powertrain efficiencies, as well as data-driven approaches to capture human factors, and uncertainty in maps and measurements, (2) efforts to address scalability and generalization of the learning in geographical areas with limited data and reduced expert supervision, and (3) the experimental validation of the platform on real-world driving data collected in a set of representative conditions. Statistical learning theory will be merged with predictive control theory using a mix of physics-based and data-driven models in the learning process. Scalability and accuracy will be attained by updating models in real-time using data and sharing models among vehicles of the same manufacturer.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.