Model-based Optimal Control of a Hybrid Electric Vehicle Using Stochastic Dynamic Programming

Deducing the optimal power split in a hybrid electric vehicle is not simple, as the solution depends on the future driving cycle, which is not usually known. Nevertheless several algorithms have been proposed to facilitate optimality in the controller design in an effort to move away from heuristic control strategies. Stochastic Dynamic Programming is used in this paper which, like other techniques proposed in literature, relies heavily on model-based design techniques to design a controller response based on a discrete set of vehicle states. Controllers based on SDP have shown promise in simulation, however real-time implementations have not been widely demonstrated. This paper presents lessons learned from the implementation of a controller for real-world use and highlights difficulties in the process, such as achieving a charge-sustaining strategy, hardware memory restrictions, and the consequence of limitations in the stochastic model. A detailed bespoke model was designed, representing a Ford Transit van equipped with a retrofit parallel (torque-addition) hybrid system. Real-world driving data were used to construct the stochastic drive cycle model, of which the LA92 test cycle is a good approximation and so was used for chassis dynamometer testing. Results showed that the model predicted baseline vehicle emissions very well, and during chassis dynamometer testing the hybrid controller was found to respond as in simulation. The SDP strategy was found to behave robustly and predictably, and to manage the battery state of charge well.