Joint online estimation of state of charge and internal temperature of lithium-ion batteries with multi-task learning

Abstract

An accurate online estimation of state of charge (SOC) is the cornerstone for a safe, efficient usage of the battery system, as well as vigilant thermal management, which relies critically on accurate temperature monitoring at locations that best reflect the thermal state. However, since temperature sensors are usually only installed on cell surfaces for reliability and safety reasons, the more informative internal temperature must be inferred indirectly from the available surface measurements. Meanwhile, the heat generation and SOC of a battery cell are often correlated with each other. Therefore, in this work, we place temperature sensors in the cell core area and cast the core temperature as state of internal temperature (SOIT) as a representative example for internal temperature monitoring of cylindrical lithium-ion cells, and propose a multi-task learning (MTL) framework, namely multi-scale multi-gate mixture-of-experts (MS-MMoE), utilizing temporal convolutional networks (TCNs) with different receptive fields (RFs) to capture the battery dynamics from different time scales for an accurate joint online estimation of SOC and SOIT. Detailed experimental results are presented under various testing scenarios, where the proposed MS-MMoE model outperforms the other MTL and single-task models in most cases and demonstrates sufficient capability to generalize on unseen cells, thereby verifying its industrial applicability. Considering the increasing demand for accurate, versatile battery state monitoring under realistic operating conditions, as well as the limited availability of internal temperature measurements, this work will support the development of next-generation intelligent BMS.