The accurate estimation of battery states like State of Charge (SOC) is a crucial task for the battery management system (BMS) of a battery electric vehicle (BEV) since safe and efficient vehicle operation depends on this. To assure a reliable estimation means to assess and adjust the algorithms responsible for the estimation under real driving conditions. Readily available driving profiles like the WLTP and profiles generated in previous works are designed with a focus on fuel consumption and energy efficiency and therefore not suited well enough for this purpose. Furthermore, edge cases in driving behavior like highly dynamic driving or constant high velocity driving are not always sufficiently covered by these profiles. For the assessment therefore, modular driving profiles under different driving conditions need to be created to simulate the respective algorithms and evaluate the results.
This work introduces a toolchain for driving profile generation of BEVs. Profiles are created via two methods: the Markov chain method utilizing the Markov chain theory and the microtrip method, with mictrotrips being short segments of a driving profile. The response of the diagnostic algorithms to the generated profiles can be evaluated by the simulation of an existing two-stage toolchain. The first stage simulation includes the vehicle dynamics during execution of the driving profile. It outputs the power profile by calculating the necessary power that needs to be delivered by the battery over time to imitate the input driving profile. In the second stage simulation the power profile is used as an input scenario and the BMS with various algorithms for SOC estimation is simulated under real-life conditions to assess the algorithm performance.
Results show that, as expected, the estimation performance can vary greatly depending on the input profile and therefore various power profiles should be considered during algorithm development to obtain a thorough understanding of the algorithm’s behavior on various driving scenarios. Specifically, it was observed that the RMSE of the SOC of the same algorithm would increase by up to 8 %pt. due to the application of different input profiles.