Data-Driven Statistical Analysis and Diagnosis of Networked Battery Systems

This paper analyzes system-level capacity, stored energy, and power loss, and introduces methods for their diagnosis with reliability characterization, for networked battery systems. Since system-level features of a networked battery system are functions of its member battery packs, it is of essential importance to establish rigorously this component-to-system relationship. The paper provides a method to calculate accurately diagnosis error probabilities on the system-level quantities when individual batteries' measurements are subject to measurement errors and their characterizing parameters are estimated from such measurements. These results are then used to choose diagnosis decision variables such as the thresholds for detecting faulty batteries and data sizes for estimation. Focusing on stored energy, capacity, and power loss, this paper characterizes error statistics of estimation algorithms for these quantities. Convergence of the algorithms, asymptotic probability distributions of the estimates, and diagnosis reliability analysis are performed rigorously by using stochastic differential equations, central limit theorems, and large deviations principles. Simulated case studies and experimental data are used to illustrate the methods and their usefulness.