Research into reduced-order models (ROM) for lithium-ion batteries is motivated by need for a real-time embedded model with the accuracy of physics-based models but having computational simplicity comparable to that of equivalent-circuit models. An attractive approach to this problem was recently proposed by Lee et al [1] with a model order reduction method known as Discrete-time-Realization Algorithm (DRA). The ROM obtained in standard state-space representation could then be used to obtain the time-evolution of all the internal electrochemical quantities of the standard porous-electrode model. An unresolved issue of this reduced-order modelling approach is the high computation requirement associated with the DRA, which has to be performed multiple times to identify cell parameters at various SOC and temperatures.

In this presentation, we analyse the computational bottleneck in the existing DRA and propose a significant improvement to the overall modelling procedure. Our analysis indicates the Singular Value Decomposition of a large Block-Hankel matrix of the system’s Markov parameters is a key inefficient step for the overall high computational requirement of DRA. A fast computational approach is presented that significantly reduces both the memory and the floating-point operation count by bypassing the redundant step of forming the large Block-Hankel matrix. Comparisons with the existing DRA method highlight the significant reduction in computation time and memory usage (Figure 1), as well as the accuracy improvement in key electrochemical quantities, afforded by our new method. With the aid of the improved modelling procedure, the transfer functions forming the porous electrode model is reformulated and suitably combined to arrive at a physics-informed equivalent circuit model of the Panasonic NCA 18650 BD cells used in automotive applications.