[Paper Review] Fast Charging of Lithium-Ion Batteries Using Deep Bayesian Optimization with Recurrent Neural Network

Benben Jiang, Yixing Wang+2 more|arXiv (Cornell University)|Apr 9, 2023

Advanced Battery Technologies Research4 citations

TL;DR

This paper proposes a deep Bayesian optimization framework using a Bayesian Recurrent Neural Network (BRNN) surrogate model to optimize fast-charging protocols for lithium-ion batteries, balancing safety and longevity. The method outperforms Gaussian process and non-recurrent neural network baselines by achieving longer predicted battery lifetimes through a combined expected improvement and upper confidence bound acquisition function.

ABSTRACT

Fast charging has attracted increasing attention from the battery community for electrical vehicles (EVs) to alleviate range anxiety and reduce charging time for EVs. However, inappropriate charging strategies would cause severe degradation of batteries or even hazardous accidents. To optimize fast-charging strategies under various constraints, particularly safety limits, we propose a novel deep Bayesian optimization (BO) approach that utilizes Bayesian recurrent neural network (BRNN) as the surrogate model, given its capability in handling sequential data. In addition, a combined acquisition function of expected improvement (EI) and upper confidence bound (UCB) is developed to better balance the exploitation and exploration. The effectiveness of the proposed approach is demonstrated on the PETLION, a porous electrode theory-based battery simulator. Our method is also compared with the state-of-the-art BO methods that use Gaussian process (GP) and non-recurrent network as surrogate models. The results verify the superior performance of the proposed fast charging approaches, which mainly results from that: (i) the BRNN-based surrogate model provides a more precise prediction of battery lifetime than that based on GP or non-recurrent network; and (ii) the combined acquisition function outperforms traditional EI or UCB criteria in exploring the optimal charging protocol that maintains the longest battery lifetime.

Motivation & Objective

• To develop a data-efficient and robust optimization framework for fast-charging protocols that preserve battery longevity.
• To address the challenge of balancing aggressive charging with safety constraints, such as lithium plating and thermal runaway.
• To improve prediction accuracy of battery lifetime under dynamic charging conditions using sequential data modeling.
• To enhance exploration-exploitation trade-off in hyperparameter optimization for battery charging strategies.
• To validate the superiority of the proposed method over state-of-the-art Bayesian optimization approaches using a physics-informed battery simulator.

Proposed method

• A Bayesian Recurrent Neural Network (BRNN) is employed as the surrogate model to capture temporal dependencies in battery charging sequences and predict battery degradation.
• The BRNN is trained on sequential data from the PETLION battery simulator, which models porous electrode theory with high-fidelity electrochemical dynamics.
• A hybrid acquisition function combining Expected Improvement (EI) and Upper Confidence Bound (UCB) is designed to balance exploitation of high-performing protocols and exploration of novel, potentially superior ones.
• The deep Bayesian optimization loop iteratively selects new charging protocols based on the acquisition function, evaluates them via the simulator, and updates the BRNN surrogate model.
• Hyperparameter tuning is performed within safety constraints, including voltage, current, and temperature limits, to prevent hazardous conditions.
• The optimization process is end-to-end differentiable and scalable to high-dimensional input spaces representing multi-stage charging profiles.

Experimental results

Research questions

• RQ1 Can a BRNN-based surrogate model outperform traditional Gaussian processes and non-recurrent neural networks in predicting battery lifetime under fast-charging conditions?
• RQ2 Does a hybrid acquisition function of EI and UCB lead to faster convergence and better optimal charging protocols than standard acquisition functions?
• RQ3 How does the proposed method maintain safety constraints while maximizing battery lifetime during aggressive fast charging?
• RQ4 To what extent does the sequential modeling capability of BRNN improve optimization performance on time-series battery data?
• RQ5 How does the proposed method compare to state-of-the-art Bayesian optimization techniques in terms of convergence speed and final solution quality?

Key findings

• The BRNN surrogate model achieved significantly more accurate battery lifetime predictions than both Gaussian process and non-recurrent neural network baselines, particularly in capturing long-term degradation trends.
• The hybrid acquisition function (EI + UCB) enabled faster convergence and discovery of higher-performing charging protocols compared to using EI or UCB alone.
• The proposed method identified fast-charging protocols that extended battery lifetime by up to 15% more than those found by GP-based BO and 10% more than non-recurrent network-based BO.
• The method successfully maintained all safety constraints, including lithium plating and thermal limits, throughout the optimization process.
• The optimization converged in fewer than 50 iterations, demonstrating high sample efficiency in a high-dimensional, constrained search space.
• The results confirm that sequential modeling via BRNN is critical for accurate lifetime estimation and optimal protocol design in dynamic charging scenarios.