A comparative study of data-driven electro-thermal models for reconfigurable lithium-ion batteries in real-time applications

Reconfigurable battery systems can change the cell topology, which allows novel applications that are not possible with conventional battery systems. However, reconfiguration complicates monitoring of the individual cell state since greater variances in temperature and SOC are introduced. Especially critical for a safe operation and adequate performance is the thermal behavior, which needs to be determined for every single cell in real-time. In this work, four different reduced-order lumped models and four artificial neural network models are compared in a surrogate modeling approach for the practical application in a reconfigurable battery system. The comparison includes the representation of serial and serial–parallel topologies. Furthermore, different coupling variants of single cell models are investigated in order to find the best representation of the physical model of a reconfigurable battery systems in an automotive application. While a separate observation of individual cells in switching systems is shown to be no longer sufficient, good results are already evident with simple models and adequate coupling approaches. The data-driven surrogate approaches achieved errors as low as 0.1 K while at the same time reducing the computational effort drastically compared to a physics-based model. •Novel application of intelligent battery systems with in-use reconfiguration.•Comparative study with 8 different approaches of electro-thermal modeling.•Model design for control-oriented modeling in real-time.•Direct comparison of reduced order models and artificial neural networks.•Qualitative and quantitative evaluation of performance and implementation.