During the operation of electric vehicles, accurate temperature monitoring of lithium-ion batteries by battery management system is crucial to their safety. However, temperature sensors possess limitations as they necessitate regular calibration to avoid inaccuracies, which can be challenging to accomplish in certain scenarios. As an alternative, a general sensorless temperature estimation framework is developed in this study to either serve as a complementary approach alongside sensors or function independently to ensure reliable temperature measurement. The proposed framework relies solely on data regarding battery voltage and current measurements, and this information is employed to determine the parameters of the equivalent circuit model via the vector-type recursive least squares algorithm in the first step. In the second step, these parameters and the voltage/current data serve as the input to the deep neural network (DNN) to estimate the instantaneous battery surface temperature during dynamic driving cycles under various ambient temperatures and degradation scenarios. To reflect the effect of past battery operation on the current temperature, a newly defined current integral function is proposed, by which the input time length of the DNN model is suggested. Four combinations of the model input are studied, and the best input option yields an average estimation error of the battery temperature that falls below 0.5 °C.
Read full abstract