Noise estimation and bias correction in identification of battery models

Battery systems demonstrate different characteristics. Their dynamics change with time and operating conditions due to aging and variations in operational conditions such as thermal distribution and and chemical properties. To enhance reliability, efficiency, and life expectancy, battery management systems must acquire high fidelity battery cell models during the course of life-time operations. This paper aims to develop identification algorithms that capture individualized characteristics of each battery cell and produce updated models in real time. It is shown that standard least-squares estimation methods will encounter identification bias. Bias correction requires information on noise statistics, which are usually unknown and time varying. This paper devises integrated algorithms for noise estimation and bias correction to resolve these issues. Noise estimation accuracy, interaction of noise estimation with signal variations, algorithm convergence, and bias correction mechanisms are discussed. A typical battery model structure is used to illustrate utilities of the methods.