Aging Responsive State of Charge Prediction of Lithium-Ion Battery Using Attention Mechanism Based Convolutional Neural Networks

Accurate estimation of State-of-Charge (SoC) in lithium-ion batteries is pivotal for their efficient energy utilization, particularly amidst aging-induced capacity decline. This paper introduces an attention-based mechanism to address this critical interplay. Integrated into a convolutional neural network framework, its efficacy is demonstrated in achieving accurate and computationally efficient co-estimation of SoC and battery capacity. A detailed comparison of the proposed algorithm with state-of-the-art methods is presented across diverse operating conditions and battery-age test case scenarios. The proposed approach enhances SoC prediction by accounting for the dynamic effects of battery aging by deploying time-ahead capacity prediction, which in turn facilitates feature correction of the SoC-prediction network. This advancement holds significant promise in optimizing the performance and longevity of lithium-ion batteries, crucial components in numerous applications ranging from portable electronics to electric vehicles and grid-scale energy storage. The algorithm has been verified on the SPEEDGOAT baseline real-time hardware-in-loop (HIL) platform. By examining the capacity degradation's influence on SoC estimation, the proposed method presents a practical and adaptable solution for actual battery management systems, significantly enhancing their overall effectiveness and reliability.