Model-based State-of-Charge Estimation of Tesla Model 3 Battery using Extended Kalman Filter

Retrieving an accurate battery's state-of-charge (SOC) is crucial for reliable operation in a variety of use cases as the era of vehicle electrification advances. Traditional methods of state-of-charge estimation often struggle with the underlying non-linear behavior and the battery's temperature effect on the estimation's accuracy. This work suggests a model-based method using the Extended Kalman Filter to estimate the battery's state-of-charge under multiple tests while accounting for the temperature effect to improve its robustness and accuracy. The testing dataset from the Tesla Model 3 NCA Li-ion battery, which McMaster University made available online, supports the study. The methodology starts with developing a second-order continuous dual-polarization (DP) equivalent circuit model. This model is used to estimate ECM parameters based on various characteristic tests and on multiple test temperatures (-20, -10, 0, 10, 25, and 40°C) given by the dataset. Once all parameters are acquired and validated, the discretized model is then developed on the base of its continuous counterpart. The EKF algorithm is then employed on top of the discretized DP model. The root mean square errors between the measured SOC and simulated SOC are calculated to quantify the performance of the EKF across multiple driving cycles. The work demonstrates a significant increase in ECM accuracy (up to 66.3% on the HWGRADE1 driving cycle) by incorporating the real-time battery temperature data into the model. The proposed EKF-based SOC estimation achieves excellent accuracy and robustness at higher temperatures with an acceptable limited error band. The EKF, however, couldn't perform adequately at low temperatures (-20°C and -10°C) due to the increasing hysteresis phenomenon which is uncaptured by the ECM model. Several solutions are proposed to tackle the problem, leaving the work open for further improvement.