Balancing Awareness Fast Charging Control for Lithium-Ion Battery Pack Using Deep Reinforcement Learning
Yang Yi-kun, Jiarui He, Chunlin Chen, Jingwen Wei
Abstract
Minimizing charging time without damaging the batteries is significantly crucial for the large-scale penetration of electric vehicles. However, charging inconsistency caused by inevitable manufacture and usage inconsistencies can lead to lower efficiency, capacity, and shorter durability due to the “cask effect.” This goal can be achieved by solving a series of constrained optimization problems with the model-based framework. Nevertheless, the high computational complexity, identifiability, and observability issues still limit their fidelity and robustness. To overcome these limitations and provide end-to-end learning strategies, this article proposes a balancing-aware fast-charging control framework based on deep reinforcement learning. In particular, a cell-to-pack equalization topology is first introduced to dispatch energy among in-pack cells. Then, the balancing awareness fast charging problem is formulated as a multiobjective optimization problem by considering charging time, consistency, and over-voltage safety constraints. Further, a deep reinforcement learning framework using a deep Q-network is established to find the optimal policy. By using the generalization of a neural network, the learned policies can be transferred to real-time charging and balancing control, thus improving the applicability of the proposed strategy. Finally, numerous comparative simulations and experimental results illustrate its effectiveness and superiority in terms of charging rapidity and balancing.