Scholarlink

<p>For batteries made of lithium-ion (Li-ion) to be more efficient, dependable, and long-lasting in applications like electric vehicles (EVs), portable devices and renewable energy storage, quick and secure charging techniques are essential. One of the challenges is the accurate estimation of internal resistance at different temperatures, loads, and under aging conditions, since the inaccuracies of estimation may compromise the charging efficiency and the safety of operation. To overcome these drawbacks, this manuscript proposes a fast and safe charging strategy of Li-ion batteries using the MSTGAN-GJO approach. The battery dataset provides the input data. Following that, the data is entered into pre-processing. In the pre-processing segment, it cleans, removes the noise and inconsistencies in the data utilizing the Maximum Correntropy Unbiased Minimum-Variance Filter (MCUMVF). The pre-processed output was fed to feature extraction using Spatio-Temporal Embedding Fusion Transformer (STEFT), which has extracted the battery electrical degradation features. After extraction, the output was fed to the Multimodal Spatial–Temporal Graph Attention Network (MSTGAN). Using MSTGAN, the battery’s internal resistance and State of Health (SOH) can be accurately predicted. Golden Jackal Optimization (GJO) is employed to maximize the weight parameter of MSTGAN. The suggested method can be implemented into use in MATLAB and evaluated using several performance metrics, such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), accuracy, precision, recall, F1-score, specificity, and sensitivity. Findings reveal that the MSTGAN-GJO approach performs better compared to other existing methods, including Deep Reinforcement Learning (DRL), Convolutional Neural Network (CNN), Reinforcement Learning (RL), Improved Gray Wolf Algorithm-Back Propagation (IGWO-BP), and Differential Search Algorithm optimized Random Forest Regression (DSA-RFR). The suggested MSTGAN-GJO has an accuracy of 0.98 with a minimum MSE of almost 0.05, which shows that it is effective and reliable in Li-ion battery applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An intelligent framework for fast and safe lithium-ion battery charging using continuous resistance estimation with multimodal spatial-temporal graph attention network and golden jackal optimization

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • K. Ramanujum,
  • K. K. Saravanan

摘要

For batteries made of lithium-ion (Li-ion) to be more efficient, dependable, and long-lasting in applications like electric vehicles (EVs), portable devices and renewable energy storage, quick and secure charging techniques are essential. One of the challenges is the accurate estimation of internal resistance at different temperatures, loads, and under aging conditions, since the inaccuracies of estimation may compromise the charging efficiency and the safety of operation. To overcome these drawbacks, this manuscript proposes a fast and safe charging strategy of Li-ion batteries using the MSTGAN-GJO approach. The battery dataset provides the input data. Following that, the data is entered into pre-processing. In the pre-processing segment, it cleans, removes the noise and inconsistencies in the data utilizing the Maximum Correntropy Unbiased Minimum-Variance Filter (MCUMVF). The pre-processed output was fed to feature extraction using Spatio-Temporal Embedding Fusion Transformer (STEFT), which has extracted the battery electrical degradation features. After extraction, the output was fed to the Multimodal Spatial–Temporal Graph Attention Network (MSTGAN). Using MSTGAN, the battery’s internal resistance and State of Health (SOH) can be accurately predicted. Golden Jackal Optimization (GJO) is employed to maximize the weight parameter of MSTGAN. The suggested method can be implemented into use in MATLAB and evaluated using several performance metrics, such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), accuracy, precision, recall, F1-score, specificity, and sensitivity. Findings reveal that the MSTGAN-GJO approach performs better compared to other existing methods, including Deep Reinforcement Learning (DRL), Convolutional Neural Network (CNN), Reinforcement Learning (RL), Improved Gray Wolf Algorithm-Back Propagation (IGWO-BP), and Differential Search Algorithm optimized Random Forest Regression (DSA-RFR). The suggested MSTGAN-GJO has an accuracy of 0.98 with a minimum MSE of almost 0.05, which shows that it is effective and reliable in Li-ion battery applications.