Optimal Model Reduction of Lithium-Ion Battery Systems Using Particle Swarm Optimization

Abstract

Lithium-ion batteries (LIBs) have been widely used as an energy storage mechanism among all the types of rechargeable batteries owing to their high energy and power density. Because of the vast applications of LIBs in several dynamic operations, the development of a robust model to simulate the battery’s dynamic behavior and performance for control and system design is paramount. Several modeling efforts have been invested into the development of electrochemical models for simulation of LIB systems ranging from a full-order model, the so-called Doyle-Fuller-Newman (DFN) model to several reduced-order models. This thesis work involves the development of a reduced-order electrochemical model based on single particle approach with electrolyte dynamics (SPMe). The partial differential equations (PDEs) that capture the dynamic behavior and performance characteristics of the LIB systems were solved numerically through a finite difference method in MATLAB environment. For model reduction purpose, a constrained optimization problem was formulated to determine the optimal uneven discretization node points needed to numerically solve the battery PDEs for both solid and electrolyte phase concentration predictions. The optimization problem was solved using a particle swarm optimization (PSO) by minimizing the errors between the reference model, a SPMe with even discretization and the reduced model, a SPMe with uneven discretization. The proposed approach is similar to that proposed by Lee T.K. and Filipi Z., but differs because of the inclusion of electrolyte dynamics. The battery voltage was computed based on the optimal uneven discretization nodes under three different charging/discharging conditions. The proposed model demonstrates that as the number of optimal uneven discretization nodes applied to the model increases, the fidelity of the model increase. However, no significant improvement of prediction accuracy is observed after a certain level of uneven discretization. The proposed model demonstrates that in comparison to the evenly discretized model, the complexity in terms of the number of states can be reduced by 7 times without loss of physical interpretation of the diffusion and migration dynamics in the solid particles and electrolyte. This reduction in the number of discretization allows for faster computation for the purpose of control and system design.