Advanced Monitoring and Real-Time State of Temperature Prediction in Lithium-Ion Cells Under Abusive Discharge Conditions Using Data-Driven Modelling

Abstract

Accurately predicting lithium-ion batteries’ state of temperature (SOT) is crucial for effective battery safety and health management. This study introduces a novel approach to SOT prediction based on voltage and temperature profiles during the abusive discharging process, aiming for enhanced prediction accuracy and evaluating the safety range. The duration of equal voltage discharge and temperature variation during discharge are considered temperature indicators. Linear regression and R2 analyses are employed to assess the relationship and variance over different discharge–charge cycles of varied duration between the complete life cycle and its temperature variance. In this study, a decision tree (DT) and an artificial neural network (ANN) are employed to estimate the SOT of a Li-ion battery. The effectiveness and accuracy of the proposed methods are validated using ageing data from eVTOL charge–discharge cycles through numerical simulations. The results demonstrate that for the short cruise range of 600 s, the DT algorithm with an R2 regression value of 6.17% demonstrates better performance than ANN, whereas for the bigger cruise range of 1000 s, the ANN model with an R2 regression value of 5.06 percent was better suited than DT. It is concluded that both DT and ANN outperform other methods in predicting the SOT of lithium-ion batteries.

Keywords

• lithium-ion battery system
• extended cruise run
• short cruise run
• linear fit regression
• decision tree
• artificial neural network