(587c) A Data-Driven Model Predictive Control Framework for Optimal Charging of Li-Ion Batteries with Experimental Validation

The growing market of electric vehicles has boosted the demand for Li-ion batteries, owing to their high energy storage capacity and power density. For the large-scale commercialization of electric vehicles, it is necessary to store and handle energy safely, reliably, and economically using Li-ion batteries. High maintenance costs and long charging times of Li-ion batteries are two main concerns for electric vehicles [1]. When a battery operates for multiple cycles of charging and discharging, its performance degrades due to various factors such as temperature change, dendrite formation, electrolytic decomposition, and mechanical stress. This degradation in performance decreases the reliability of a battery, eventually affecting its lifetime. In addition, although fast charging of a battery is desired, increasing charging rates to achieve fast charging can lead to accelerated capacity degradation which increases safety concerns and reduces battery lifetime. Hence, for a safe and economically desirable battery operation, it is crucial to implement a charging strategy that minimizes the charging time of a battery while maximizing its lifetime.

Currently, model-free approaches based on pre-defined voltage and current limits are mainly used for determining charging protocols [2]. These methods are executed through heuristic determination of charging parameters, thus limiting their ability to optimize a battery’s performance [3, 4]. In contrast, model-based approaches are considered to have a high potential for determining an optimal charging strategy, especially when data-driven battery models are utilized. Motivated by this, we develop a model predictive control (MPC) framework using sparse identification of nonlinear dynamics (SINDy). The SINDy algorithm is a sparse modeling technique that captures the underlying nonlinear process dynamics from historical operational data [5]. In the developed method, we predict the future behavior of a battery through SINDy. Specifically, we employ two SINDy models to describe two-timescale battery dynamics: inter-SINDy to predict capacity degradation at the end of every operating cycle and intra-SINDy to predict the evolution of state of charge (SoC) and voltage within each operating cycle. Capacity degradation indicates the decreasing lifetime of a battery, and SoC and voltage provide estimation of charging time. These predictions are used to solve an optimization problem to meet the competing control objectives of maximizing lifetime and minimizing charging time while satisfying any operational constraints. As a result, we obtain an optimal charging profile. The developed MPC framework is demonstrated and validated using experimental data for an NMC 811 battery with cathode comprising of 80 % Nickel, 10 % Manganese, and 10 % Cobalt and Li-ion inserted in carbon nanotubes as anode.

Literature cited:

1. Ouyang, Q., Wang, Z., Liu, K., Xu, G. and Li, Y., 2019. Optimal charging control for lithium-ion battery packs: A distributed average tracking approach. IEEE Transactions on Industrial Informatics, 16(5), pp.3430-3438.
2. Yin, Y., Bi, Y., Hu, Y. and Choe, S.Y., 2021. Optimal fast charging method for a large-format lithium-ion battery based on nonlinear model predictive control and reduced order electrochemical model. Journal of The Electrochemical Society, 167(16), p.160559.
3. Tian, N., Fang, H. and Wang, Y., 2020. Real-time optimal lithium-ion battery charging based on explicit model predictive control. IEEE Transactions on Industrial Informatics, 17(2), pp.1318-1330.
4. Zou, C., Manzie, C. and Nešić, D., 2018. Model predictive control for lithium-ion battery optimal charging. IEEE/ASME Transactions on Mechatronics, 23(2), pp.947-957.
5. Brunton, S.L., Proctor, J.L. and Kutz, J.N., 2016. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the national academy of sciences, 113(15), pp.3932-3937.