Online Prediction of Temperature Distribution in Power Batteries Under Sparse Sensing: An Eigenmode Approximation Perspective

Monitoring the global temperature distribution in Lithium-ion power batteries is essential for refined thermal management. However, nonlinear dynamics and the limited usage of sensors poses considerable challenges for model-driven temperature regulation. This study introduces an eigenmode approximation-based data-driven modeling framework for online predicting full-state temperature fields under sparse sensing. The framework comprises offline training, sensor optimization, and online learning. The proper orthogonal decomposition is initially employed to extract typical eigenmodes of temperature fields from offline data. Subsequently, the orthogonal-triangular (QR) factorization with column pivoting (QR-CP) algorithm is used to optimize sensor locations approximately and identify low-order representations of temperature fields. On this basis, a machine learning model is built to track unknown system states online and reconstruct the temperature field through predictive results. Experimental studies demonstrate the proposed method’s reliable performance and sensor reduction capabilities.