Scholarlink

Addressing the urgent need for accurate State of Health (SOH) prediction in lithium-ion batteries to mitigate safety risks such as thermal runaway, this study proposes an innovative TPE-BiLSTM-Attention hybrid algorithm based on NASA’s standard battery dataset. The algorithm deeply integrates three synergistic capabilities: the efficient global hyperparameter optimization of Tree-structured Parzen Estimator (TPE) Bayesian optimization, the bidirectional degradation feature modeling of Bidirectional Long Short-Term Memory (BiLSTM) networks, and the dynamic local feature focusing of attention mechanisms, constructing a high-performance predictive model. Using TPE optimization, key hyperparameters of the BiLSTM-Attention model are globally optimized, significantly enhancing SOH prediction accuracy and robustness. Experimental results demonstrate the algorithm’s outstanding performance in SOH estimation, achieving a root mean square error (RMSE) of 0.9232% and a mean square error (MSE) of 0.8523%.

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

TPE Bayesian-Optimized Attention-Enhanced Bidirectional LSTM Model and Its Application in Lithium-Ion Battery Lifespan Prediction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • ZhenRong Yuan,
  • KeFeng Huang,
  • CaiHua Xu,
  • JunChao Zou,
  • Jun Yan

摘要

Addressing the urgent need for accurate State of Health (SOH) prediction in lithium-ion batteries to mitigate safety risks such as thermal runaway, this study proposes an innovative TPE-BiLSTM-Attention hybrid algorithm based on NASA’s standard battery dataset. The algorithm deeply integrates three synergistic capabilities: the efficient global hyperparameter optimization of Tree-structured Parzen Estimator (TPE) Bayesian optimization, the bidirectional degradation feature modeling of Bidirectional Long Short-Term Memory (BiLSTM) networks, and the dynamic local feature focusing of attention mechanisms, constructing a high-performance predictive model. Using TPE optimization, key hyperparameters of the BiLSTM-Attention model are globally optimized, significantly enhancing SOH prediction accuracy and robustness. Experimental results demonstrate the algorithm’s outstanding performance in SOH estimation, achieving a root mean square error (RMSE) of 0.9232% and a mean square error (MSE) of 0.8523%.