Interpretable and sensorless lithium-ion battery temperature estimation: A Temporal Fusion Transformer approach

Abstract

Accurate estimation of surface temperature in lithium-ion batteries is crucial for safety, performance, and longevity in applications. In this study, we propose a sensorless and interpretable lithium-ion battery surface temperature estimation model based on Temporal Fusion Transformer (TFT). Our model utilizes physical quantities derived from the analysis of the battery's equivalent circuit model, a simplified thermal model, and the electrochemical reaction process as inputs for temperature estimation. The advanced TFT model, equipped with a multi-head interpretable attention mechanism, is employed for accurate temperature estimation. Experimental results demonstrate that our proposed model achieves maximum temperature errors below 1 °C under various driving conditions and fluctuating ambient temperatures (−10 °C to 20 °C), meeting the expected outcomes. Furthermore, we enhance the interpretability of the model by variable-based ante-hoc and post-hoc strategy. This methodology provides a dependable and economical substitute for conventional sensor-based methods in battery management systems, offering implications for improved battery performance and safety.