In the quest to optimize battery charging protocols, researchers have often faced a daunting challenge: each evaluation is slow, costly, and non-differentiable. This has led to approaches that heavily constrain the protocol search space, limiting the diversity of protocols that can be explored and potentially preventing the discovery of higher-performing solutions. A team of researchers from MIT’s Laboratory for Information and Decision Systems, including Ge Lei, Ferran Brosa Planella, Sterling G. Baird, and Samuel J. Cooper, have introduced a novel solution to this problem using large language models (LLMs).
The researchers have developed two gradient-free, LLM-driven closed-loop methods: Prompt-to-Optimizer (P2O) and Prompt-to-Protocol (P2P). P2O uses an LLM to propose the code for small neural-network-based protocols, which are then trained by an inner loop. P2P, on the other hand, simply writes an explicit function for the current and its scalar parameters. Both methods aim to expand the space of protocol functional forms, incorporate language-based constraints, and enable efficient optimization in high-cost experimental settings.
In their case studies, the researchers found that LLM-guided P2O outperformed neural networks designed by Bayesian optimization, evolutionary algorithms, and random search. In a realistic fast charging scenario, both P2O and P2P yielded around a 4.2 percent improvement in state of health (a capacity retention-based health metric under fast charging cycling) over a state-of-the-art multi-step constant current (CC) baseline. Notably, P2P achieved this under matched evaluation budgets, meaning the same number of protocol evaluations.
The practical applications for the energy sector are significant. Faster, more efficient charging protocols could lead to reduced charging times for electric vehicles, increased battery lifespan, and improved overall performance. This research, published in the journal Nature Communications, demonstrates the potential of LLMs in optimizing complex processes in the energy industry.
This article is based on research available at arXiv.