Scholarlink

It is essential to predict the remaining useful life (RUL) of a lithium-ion battery for the reliability and safety operation of an energy storage system. In this direction, this paper proposes an integrated battery prognostic methodology combining principal component analysis (PCA) and long short-term memory (LSTM) networks for the RUL prediction of a battery. Firstly, a robust preprocessing pipeline employing sliding window anomaly detection for outlier removal and the K-Nearest Neighbors(KNN) algorithm for missing data refilling is developed. Secondly, the battery capacity is accurately extracted through the Ampere-hour integration method, while PCA is employed to construct the health indicators (HIs) by dimensionality reduction of eight degradation features, including cycle number, internal resistance, and charge-discharge characteristics. These HIs serve as the inputs to a customized Long Short-Term Memory (LSTM) network to predict the remaining capacity trajectories. Numerically experimental results demonstrate the superior performance of the proposed methodology over other algorithms.

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

A PCA-LSTM Methodology for Lithium-Ion Battery Remaining Useful Life Prediction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Dingmei Wang,
  • Pengfei Gao,
  • Jinping Zhang,
  • Jin Li,
  • Haozhe Zhang,
  • Leibang Liu

摘要

It is essential to predict the remaining useful life (RUL) of a lithium-ion battery for the reliability and safety operation of an energy storage system. In this direction, this paper proposes an integrated battery prognostic methodology combining principal component analysis (PCA) and long short-term memory (LSTM) networks for the RUL prediction of a battery. Firstly, a robust preprocessing pipeline employing sliding window anomaly detection for outlier removal and the K-Nearest Neighbors(KNN) algorithm for missing data refilling is developed. Secondly, the battery capacity is accurately extracted through the Ampere-hour integration method, while PCA is employed to construct the health indicators (HIs) by dimensionality reduction of eight degradation features, including cycle number, internal resistance, and charge-discharge characteristics. These HIs serve as the inputs to a customized Long Short-Term Memory (LSTM) network to predict the remaining capacity trajectories. Numerically experimental results demonstrate the superior performance of the proposed methodology over other algorithms.