Simultaneous Prediction of Three Key Li-Ion Battery Indicators through Multi-Task Learning

Accurate diagnosis of battery states is essential for ensuring operational safety, evaluating current performance, and assessing suitability for second-life applications. However, characterizing key indicators such as State of Health (SoH), Direct Current Internal Resistance (DCIR), and maximum cell temperature during operation (Max. temp.) remains highly time-intensive because it requires low-current measurements and separate testing procedures. The challenge is further compounded by the weak linearity and large variance in the relationships among these indicators. To address these challenges, we developed a multi-task learning (MTL) framework that combined a rapid diagnostic protocol with simultaneous prediction of the three key battery indicators. Using features extracted from the rapid diagnostic protocol, a neural network model was trained to predict multiple targets concurrently. The proposed MTL framework outperformed single-task baselines, reducing the root mean squared error by 25.9% for SoH, 25.0% for the DCIR increase rate, and 29.4% for Max. temp. These performance gains arise from the joint learning of highly correlated targets, which enables the model to capture coupled degradation behavior and promotes positive transfer and inductive bias. By integrating a rapid diagnostic protocol with an MTL, this study provides an accurate and data-efficient framework for simultaneous prediction of multiple degradation-relevant battery indicators.