A Data-Driven Power Management Strategy for Plug-In Hybrid Electric Vehicles Including Optimal Battery Depth of Discharging

Abstract

For hybrid electric vehicles, higher depth of discharge (DOD) indicates more use of battery energy, which can supply more inexpensive propulsions than the fossil fuel but accelerate the battery aging, thus leading to the increase in the equivalent battery life loss cost (EBLLC) related to battery aging. While developing an energy management strategy considering the battery aging effect for plug-in hybrid electric vehicles (PHEVs), a tradeoff between energy consumption cost (ECC) and EBLLC should be made to identify the optimal DOD and minimize the total cost (TC). Furthermore, the optimal DOD is changeable with the initial state of charge (SOC) level. To develop a robust controller to deal with varying initial SOCs for PHEVs, this paper proposes a data-driven method, namely, a three-layer artificial neural network (ANN) to realize real-time power distribution including battery life model. Real-world speed profiles and Pontryagin's minimum principle (PMP) are leveraged to identify the optimal DODs and generate the neural network training data based on cases with a range of initial SOCs. The results clearly demonstrate the robustness of the proposed ANN and also indicate that the data-driven method can effectively reduce the total of ECC and EBLLC compared to typical optimization algorithms without a battery aging model, including the dynamic programming, PMP, and the rule-based strategy. In particular, the ANN can reduce the TC by 19.99%, 25.97%, and 33.13%, respectively, for cases with the initial SOC of 0.95, 0.85, and 0.65, compared to the rule-based method. And the TC of the ANN is comparable to the PMP including the battery degradation model. Moreover, the training sample scale on forecasting accuracy and computational efficiency of the ANN is evaluated. Finally, the computational time of these methods is comprehensively discussed to evaluate the time efficiency of the proposed method.