Parallel Bayesian optimization using satisficing Thompson sampling for fast charging design of lithium-ion batteries

Abstract

The rapid growth of the battery industry has increased the demand for fast charging strategies, which face two key optimization challenges: (i) the large space of the candidate charging strategy compared to the limited experiment budget, and (ii) the limited understanding of various degradation mechanisms. Existing optimization methods can be broadly classified into model-based and data-driven approaches. Model-based approaches are constrained by the accuracy of degradation models, while classical data-driven methods, such as Bayesian optimization (BO), may require many iterations to find the optimal solution. In this paper, we propose a data-driven approach named Blahut-Arimoto satisficing Thompson sampling-based parallel Bayesian optimization (BLASTS-PBO) for fast charging optimization. BLASTS-PBO shifts the target from the optimal solution to a satisficing solution that balances the informational requirements and optimality, which is particularly beneficial for time-sensitive optimization problems. We employ Gaussian processes as the surrogate model to capture the relationship between different charging strategies and utilize the BLASTS algorithm to compute the satisficing strategy. We also introduce the parallel technique to further improve time efficiency. Theoretical bounds on Bayesian cumulative regret confirm the method’s efficacy, and experiments using a porous electrode theory-based battery simulator demonstrate that BLASTS-PBO outperforms both sequential counterparts and parallel BO with traditional Thompson sampling in both synchronous and asynchronous settings. The results underscore BLASTS-PBO’s practical implication in accelerating the identification of effective charging strategies, offering a valuable solution for time-sensitive battery optimization problems.