Online prediction for heat generation rate and temperature of lithium-ion battery using multi-step-ahead extended Kalman filtering

Abstract

It is well known that the performance, reliability, and safety of lithium-ion batteries (LIBs) are strongly influenced by temperature. Owing to the hysteresis of temperature regulation, multi-step-ahead online temperature prediction is needed for model prediction control of battery thermal behaviors, and for improving the safety and service life of battery. According to the energy conservation equation, temperature is determined on the account of heat dissipation and generation. In that, when facts pertaining to heat dissipation are understood and known to a certain extent, battery heat generation ought to be the key to battery temperature prediction. Nonetheless, measurement of heat generation rate based on calorimetry is unfit for online applications, and calculation of heat generation rate based on the Bernardi equation also has limitations, such as error escalation with large-size LIBs or sensitivity to the battery’s entropy coefficient. To circumvent these issues, this paper considers a prediction method for battery heat generation rate and temperature based on the multi-step-ahead extended Kalman filter, which is profoundly useful for nonlinear state estimation. The filtering process takes two iterative steps, in that, the Bernardi equation is used as the state equation, and the battery energy conservation equation is taken as the observation equation. The proposed method is able to reduce the mean absolute error (MAE) of heat generation rate prediction by about 40% as compared to the benchmark that uses Bernardi equation alone, and the MAE of temperature is less than 0.1 K under the Urban Driving Cycle working condition, over different prediction horizons. Moreover, when the entropy coefficient fluctuates within the range of 0.7 to 1.3 times of the original value, the MAE variation of heat generation rate is less than 0.04 W while that of the aforementioned benchmark is 0.13 W. In short, the proposed method can improve both the accuracy and robustness of prediction of heat generation rate, while satisfying online applications.