Mitigation of Voltage Violation for Battery Fast Charging Based on Data-Driven Optimization

Fast charging of lithium-ion batteries is a pivotal technology for diminishing charging duration and augmenting user convenience. Nevertheless, with the escalation in battery power, the battery voltage experiences a swift upsurge during fast charging. When the battery voltage surpasses the voltage threshold defined by operational limits, irreversible damage to the battery becomes inevitable. Indeed, the process of fast charging inherently represents a multi-objective optimization challenge, wherein the reduction of charging duration and the prevention of battery voltage violations stand as two opposing objectives. In this work, we introduce a multi-objective reinforcement learning algorithm aimed at devising fast-charging strategies that effectively balance the trade-off between charging duration and battery voltage violation. To begin with, we establish the Doyle-Fuller-Newman (DFN) model for lithium-ion batteries, upon which the framework for the fast-charging process is constructed. Next, the fast charging process is mathematically framed as a Markov decision process (MDP). Subsequently, we present a multi-objective reinforcement learning algorithm tailored to address the MDP problem. Lastly, the effectiveness of the devised algorithm is validated through a series of simulation experiments. The simulation results demonstrate that the introduced algorithm adeptly generates efficient fast-charging strategies for lithium-ion batteries based on the specified preferences.