An online hybrid estimation method for core temperature of Lithium-ion battery with model noise compensation

Abstract

Temperature monitoring plays an important role in developing advanced battery management systems, ensuring safety, and improving cell performance. Core temperature provides more accurate indications of battery natures than surface temperature, but it cannot be measured directly. In this paper, a novel hybrid method by fusing a model-based method and a data-driven method is proposed to estimate the battery core temperature with model noise compensation. In the model-based method, an extended Kalman filter (EKF) is developed to estimate the core temperature based on an electro-thermal coupling model. The model parameters are updated with the feedback of the estimated core temperature and state of charge. In the data-driven method, a neural network is trained to characterize the battery model noises. For model noise compensation, the noise covariances of the EKF are dynamically adjusted by minimizing the estimation errors between the EKF and the neural network with particle swarm optimization. Experiments for implementing and validating the proposed method are conducted in a wide range of ambient temperatures. Compared with three existing methods, the proposed method can improve the estimation accuracy by at least 56.8% at −15 °C and 60.9% at 5 °C.