State-of-health (SOH) estimation is crucial for ensuring efficient, reliable and safe operation of power battery in electric vehicle (EV) application. However, due to the complicated physicochemical reactions happened in battery cells, it is extremely difficult to accurately estimate SOH, especially in real-world EV application scenarios. Traditional SOH estimation methods, including both model-based and data-driven ones, are deterministic, which cannot capture the stochastic property of battery aging process aroused from the inherent inconsistency during battery production. In this paper, Bayesian network (BN), which is a probabilistic graphical modeling method for indeterministic process, is used to battery degradation modeling. Its structure is derived from existing knowledge about battery aging mechanism. Two-year operational data and capacity calibration results of 16 electric taxies are collected for model training and validation. Specifically, a systematic data filling procedure is proposed to predict the missing values of variables necessary for SOH estimation. Markov Chain Monte Carlo method is adopted to generate the samples from parameterized BN for SOH estimation. Results show that the estimation result is very close to the calibrated SOH with mean absolute error below 4%. The proposed method is promising to be applied online for SOH estimation in real-world EV application.INDEX TERMS Electric vehicle, battery aging, state-of-health estimation, real-world data.

No abstract

In this paper, an online energy management strategy (EMS) for hybrid electric tracked vehicle (HETV) is developed based on deep deterministic policy gradient (DDPG) with time-varying weighting factor to further improve economic performance of HETV and reduce computational burden. The DDPG is applied to model the EMS problem for the target HETV. Especially, a time-varying weighting factor is introduced here to update old network parameters with experience learned from most recent cycle segment. Afterwards, simulation is conducted to verify the effectiveness and adaptability of the proposed method. Results show that DDPG-based EMS with online updating mechanism can achieve nearly 90% fuel economy performance as that of dynamic programming while computational time is greatly reduced. Finally, hardware-in-loop experiment is carried out to evaluate the real-world performance of the proposed method.

No abstract