A novel joint estimation for core temperature and state of charge of lithium-ion battery based on classification approach and convolutional neural network

Abstract

The core temperature (CT) of lithium-ion batteries (LIBs) is crucial for effective thermal management, and is significantly influenced by the state of charge (SOC) among other battery parameters. However, directly measuring CT and SOC poses challenges. This study introduces a novel joint estimation method for CT and SOC, in which classification approach is pioneeringly employed through a two-dimensional convolutional neural network (2D-CNN). SOC is initially estimated and subsequently utilized as an input to refine CT estimation. This process is facilitated by two 2D-CNNs that incorporate measured indices such as current, terminal voltage, surface temperature, and ambient temperature into their inputs. Techniques including data augmentation, sequential optimization, and Gaussian filtering are employed to enhance estimation effect. The methodology is validated using LIB charge/discharge data derived from experiments based on the urban dynamometer driving schedule (UDDS). Results demonstrate that the root-mean-square errors (RMSEs) for CT and SOC estimation are below 0.267 °C and 0.7 %, respectively, across a broad ambient temperature range of −15 °C–55 °C. Consequently, this study confirms the effectiveness of using classification for CT and SOC estimation, illustrating that the proposed method can precisely estimate CT and SOC of LIBs across a wide temperature spectrum. The effectiveness of the deep learning method based on classification in estimating CT and SOC has been validated. Furthermore, a comparative analysis reveals that incorporating SOC as an input for CT estimation notably reduces the training time required by the method.