Summary Uncertainty prediction of lithium‐ion battery state‐of‐charge (SOC) is key for electric vehicle battery management systems. Aiming at the shortcomings of a single equivalent circuit model (ECM) and traditional SOC fusion estimation algorithms, this paper proposes a new multi‐model SOC fusion method. First, three sub‐models are established. Second, an adaptive extended Kalman filter is applied to each sub‐model in parallel to predict the battery terminal voltage and SOC simultaneously. Then, based on the ordered weighted averaging (OWA) operator theory, the real covariance matrix of the output voltage error of each model is obtained, and the weight factor of each sub‐model is calculated using this matrix. Finally, the SOC estimation of each model is weighted and synthesized to realize the SOC fusion estimation. The experimental results show that the maximum absolute error of the multi‐model SOC fusion estimation based on the OWA operator is close to the optimal value of a single model, whether it is the fusion of three ECMs or two ECMs with a degraded electrochemical model, and the multi‐model SOC fusion estimation based on OWA operator has better robustness than the single model SOC estimation.

SummaryThe design of energy management strategy (EMS) plays a vital role in the power performance and economy of battery–ultracapacitor for electric vehicles. A reinforcement learning (RL)‐based EMS is proposed to obtain an optimal power allocation strategy for battery–ultracapacitor electric vehicle, and its robustness is verified at different temperatures. First of all, the dynamic characteristic experiments of the battery and ultracapacitor were performed at 10°C, 25°C, and 40°C to obtain mechanism characteristics at different temperatures. Secondly, a genetic algorithm is selected to identify the parameters of the battery and ultracapacitor model. Next, the RL‐based strategy takes the minimum energy loss of the hybrid energy storage system as the reward function and solves the optimal policy based on Markov theory. The simulation results show that the economy of the RL‐based strategy correspondingly improved by 3.05%, 3.20%, and 3.15% at different temperatures in comparison with the fuzzy‐based strategy, and the economic gap between the RL‐based strategy and the DP‐based strategy is further narrowed down to 7.30%, 3.88%, and 8.40% at different temperatures, respectively. Finally, the proposed strategy is validated under different driving conditions, which indicate that the RL‐based strategy can effectively reduce energy consumption and has good robustness at different temperatures.

No abstract