Comparison of Optimal Energy Management Strategies Using Dynamic Programming, Model Predictive Control, and Constant Velocity Prediction

Due to the recent advancements in autonomous vehicle technology, future vehicle velocity predictions are becoming more robust, which allows fuel economy (FE) improvements in hybrid electric vehicles (HEVs) through optimal energy management strategies (EMS). Velocity predictions generated between 5 and 30 s predictions could be implemented using model predictive control (MPC), but the performance of MPC must be well understood. Also, the vulnerability of predictive optimal EMS to velocity prediction accuracy should be addressed. Before an optimal EMS can be implemented, its overall performance must be evaluated and benchmarked against relevant velocity prediction metrics. A real-world highway drive cycle (DC) in the high-fidelity, controls-oriented 2017 Toyota Prius Prime model operating in charge-sustaining mode was utilized to observe FE realization. We propose three important metrics for comparison to no velocity prediction control: (1) perfect full DC prediction using dynamic programming (DP), (2) 10 s prediction horizon MPC, (3) 10 s constant velocity prediction. The very first optimal EMS requires the whole DC prediction in advance, whereas the rest of the two strategies only require limited horizon velocity prediction. These different velocity predictions are input into an optimal EMS derivation algorithm to derive optimal engine torque and engine speed. Our results show that the constant velocity prediction algorithm outperformed the baseline control strategy, but underperformed the MPC strategy. We also show that using a 10 s prediction window, MPC strategy provided FE improvement results very close to the full DC prediction case. So with the advancement in perception systems, MPC could be implementable in real vehicles. Also, constant velocity prediction results are good enough that optimization should be implemented in vehicle controllers with MPC type of framework. In future work, we seek to combine these models with velocity prediction models to demonstrate FE improvements on a physical vehicle.