Fuel and CO2 savings in real driving using machine learning HEV operating strategy
Today, HEV operating strategies are optimized mainly in regards to the legal driving cycles. This leads to increased fuel consumption in customer operation, as this was not part of the optimization process. The paper shows, how learning algorithms can be used to extend a control unit applicable operation strategy, in the way that the operation parameters can be optimally fit to cycle as well as customer operation. Hence, the optimum of fuel consumption and CO2 can be achieved.
Based on an adaptive operation strategy, best parameterization is chosen by a driving style- and driving environment identifier (from the DAS-sector), using identification algorithms, which are validated by measurement data. The parameter sets for different driving style and environment combinations have previously been collected.
Thereafter, a concept to enhance the operating strategy by machine learning is proposed, which utilizes an individual adaption to the driver and achieves the global consumption optimum independent on the driving situation.
The paper displays the CO2 reductions of the learning operating strategy in comparison to a basic heuristic operating strategy. The investigations show, that in real driving conditions on average 14% CO2, can be saved using a controlled learning operating strategy.
KeywordsOperating Strategy Driving Cycle Driving Environment Parameter Pice Driving Situation
Unable to display preview. Download preview PDF.