Assessing the potential of predictive control for hybrid vehicle powertrains using stochastic dynamic programming

The potential for reduced fuel consumption of hybrid electric vehicles by the use of predictive powertrain control was assessed on measured-drive data from an urban route with varying topography. The assessment was done by evaluating the fuel consumption using three optimal controllers, each with a different level of information access to the driven route. The lowest information case represents that the vehicle knows that it is being driven in a certain environment, e.g., city driving, and that the controller has been optimized for that type of environment. The second highest information level, represents a vehicle equipped with a GPS combined with a traffic-flow information system. In the highest information level, the future power demand is completely known to the control system, hence, the corresponding optimal controller results in the minimal attainable fuel consumption. This paper showed that good performance (1%-3% from the minimal attainable fuel consumption) can be achieved with the lowest information case, with a time-invariant controller that is optimized to the environment. The second highest information level results in less than 0.2% higher consumption than the minimal attainable on the studied route. This means that it is possible to design a predictive controller based on information supplied by the vehicle-navigation system and traffic-flow-information systems that can come very close to the minimal attainable fuel consumption. A novel algorithm that uses information supplied by the vehicle-navigation system was presented. The proposed algorithm results in a consumption only 0.3% from the minimal attainable consumption on the studied route.