Cell behaviour-aware power capability prediction and operation optimization of large-scale battery systems

2026-07-22

Large-scale battery system applications, e.g., transportation electrification and grid-level renewable energy integration, are vital for the EU to achieve net-zero emissions by 2050. In such battery systems, real-time operating behaviours of individual battery cells determine the system’s overall performance and safety, highlighting the need to develop cell behaviour-aware control algorithms for advanced battery system management. Thus, in this proposal, we will take into account the specific operating limits of all internal battery cells for accurately predicting and significantly boosting battery system's maximum allowable charge and discharge power, i.e., state of power (SoP). Then, the obtained cell behaviour-aware battery system SoPs will act as the operating power limits for further optimising the system's operation performance. As an MSCA-PF fellow, Dr. Weiji Han will receive crucial training at the Chalmers University of Technology and engage in detailed battery system modelling and control works which span the areas of circuit analysis, automatic control, and optimisation. The interplay between these areas will spark an innovative and productive throughput, including 1) an accurate prediction algorithm for the long-horizon SoPs of battery systems, 2) a high-efficiency dynamic reconfiguration algorithm for boosting battery system SoPs, and 3) an optimal control framework of dynamic reconfiguration for optimising battery system performance based on adapted deep reinforcement learning (DRL). The major novelties or contributions will include: considering both battery parameter update and operating constraints during the prediction horizon to enhance the prediction accuracy of long-horizon SoPs; unifying the modelling of reconfigurable battery systems to significantly reduce the modelling efforts and facilitate the formulation of control or optimisation problems; narrowing the search space and adapting the DRL method to achieve higher computational efficiency.

Sweden

2026-01-01