Fuel Consumption Reduction Through Velocity Optimization for Light-Duty Autonomous Vehicles with Different Energy Sources

Abstract

The emergence of self-driving cars provides an additional flexibility to the vehicle controller, by eliminating the driver and allowing for control of the vehicle's velocity. This work employs constrained optimal control techniques with preview of position constraints, to derive optimal velocity trajectories in a longitudinal vehicle following mode. A framework is developed to compare autonomous driving to human driving, i.e. the Federal Test Procedures of the US Environmental Protection Agency. With just velocity smoothing, improvements by offline global optimization of up to 18% in Fuel Economy (FE), are shown for certain drive cycles in a baseline gasoline vehicle. Applying the same problem structure in an online optimal controller with 1.5 s preview showed a 12% improvement in FE. This work is further extended by using a lead velocity prediction algorithm that provides inaccurate future constraints. For a 10 s prediction horizon, a 10% improvement in FE has been shown.

A more conventional procedure for achieving velocity optimization would be the minimization of energy demand at the wheels. This method involves a non-linear model thus increasing optimization complexity and also requires additional information about the vehicle such as mass and drag coefficients. It is shown that even though tractive energy minimization has a lower energy demand than velocity smoothing, smoothing works as well if not better when it comes to reducing fuel consumption. These results are shown to be valid in simulation across three different engines ranging from 1.2 L-turbocharged to 4.3 L-naturally aspirated. The implication of these results is that tractive energy minimization requiring more complex control does not work well for conventional gasoline vehicles. It is further shown that using reduced order powertrain models currently found in literature for velocity optimization, can result in worse FE than previous optimizations. Therefore, an easily implementable, vehicle agnostic velocity smoothing algorithm could be preferred for drive cycle optimization.

Employing these same velocity optimization techniques for a battery electric vehicle (BEV) can increase battery range by 15%. It is further demonstrated that eco-driving and regenerative braking are not complimentary and eco-driving is always preferred. Finally, power split optimization has been carried out for a fuel cell hybrid, and it has been shown that a rule-based strategy with drive cycle preview could match the global optimal results.