Scholarlink

<p>Precise estimation regarding the State of Health (SOH) of lithium-ion batteries constitutes a foundational element for the secure management of energy storage systems, though its realization is obstructed by the non-linear progression of degradation and the entangled relationships in multivariate streams. To address these issues, the Exponential Smoothing Transformer was chosen as the backbone network to capture long-term degradation trends. Based on this, the TREformer framework is proposed, which introduces Temporal-Channel Coordinate Attention (TCCA) to extract positional dependencies between variables and Reversible Instance Normalization (RevIN) to counteract the inherent statistical non-stationarity in battery aging. By embedding the attention mechanism preceding the encoder, the model achieves a collaborative synthesis of long-range dependencies spanning both temporal and channel coordinates. The architecture deploys a symmetrical reversible normalization strategy to counteract statistical non-stationarity, rectifying data distribution shifts. Experiments conducted on the NASA and CALCE public datasets demonstrate that the proposed method is applicable to battery data under various charging and discharging strategies. Under all test conditions, the MAE, RMSE, and NRMSE consistently remain below 2.65%, significantly outperforming mainstream time series models and providing an effective solution for battery health management under complex operating conditions.</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

A method for estimating the SOH of lithium-ion batteries based on exponential smoothing transformer with temporal-channel coordinate attention and reversible instance normalization

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Shanshan Wang,
  • Kaiming Chen,
  • Zuhao Jin,
  • Aolin Qin,
  • Liang Zeng

摘要

Precise estimation regarding the State of Health (SOH) of lithium-ion batteries constitutes a foundational element for the secure management of energy storage systems, though its realization is obstructed by the non-linear progression of degradation and the entangled relationships in multivariate streams. To address these issues, the Exponential Smoothing Transformer was chosen as the backbone network to capture long-term degradation trends. Based on this, the TREformer framework is proposed, which introduces Temporal-Channel Coordinate Attention (TCCA) to extract positional dependencies between variables and Reversible Instance Normalization (RevIN) to counteract the inherent statistical non-stationarity in battery aging. By embedding the attention mechanism preceding the encoder, the model achieves a collaborative synthesis of long-range dependencies spanning both temporal and channel coordinates. The architecture deploys a symmetrical reversible normalization strategy to counteract statistical non-stationarity, rectifying data distribution shifts. Experiments conducted on the NASA and CALCE public datasets demonstrate that the proposed method is applicable to battery data under various charging and discharging strategies. Under all test conditions, the MAE, RMSE, and NRMSE consistently remain below 2.65%, significantly outperforming mainstream time series models and providing an effective solution for battery health management under complex operating conditions.