Learning-based sparse spatiotemporal modeling for distributed thermal processes of Lithium-ion batteries

Distributed thermal modeling of Lithium-ion batteries (LIBs) is critical for the safety of electric vehicles. Due to the installation and cost constraints, only limited sensors are allowed for practical applications. In this paper, a learning-based framework is proposed for online spatiotemporal modeling of distributed thermal processes in pouch-type LIBs under sparse sensing. It consists of two stages. In the offline learning stage under full sensing, the Karhunen-Loeve (KL) decomposition is used to extract the full spatial basis functions (BFs). In the subsequent online modeling stage under sparse sensing, a spatial mapping filter is first designed to recover the missing spatial information using the initial full BFs, which are then dynamically updated by the incremental KL technique as the thermal process evolves. By iteratively repeating these two steps, the streaming sparse spatiotemporal output can be accurately completed. Finally, the typical KL-based time-space separation method can be used for online temperature prediction. The simulation results of the distributed thermal processes on a pouch-type cell and a LIB pack demonstrate the effectiveness of the proposed method.