Scholarlink

<p>State-of-health (SOH) estimation and remaining useful life (RUL) prediction of lithium-ion batteries are of great importance in ensuring safe and reliable management of electric vehicles. However, due to the phenomena of capacity regeneration and nonlinear degradation, SOH estimation faces challenges in achieving high precision while maintaining a low model parameter count. Additionally, accurate prediction of RUL throughout the battery’s entire lifecycle is a challenge. To solve the problems, FISNet is proposed to predict SOH and the Phased-RUL Approach is proposed to estimate RUL. FISNet employs a complex-valued linear layer to learn the periodic patterns of the SOH curve and perform reasonable interpolation, specifically capturing degradation trends at low frequencies and regeneration phenomena at high frequencies. The Phased-RUL Approach employs different techniques in two phases of the battery to achieve high precision. In the first phase, it employs a data-driven approach to directly predict RUL through a hybrid model integrating convolutional neural networks, Box-Cox transformation, and multi-layer perceptron block. In the second phase, it employs FISNet to predict SOH and calculate RUL. The experimental results demonstrate that FISNet significantly reduces the model parameter count to 1/250 of the average of four current models and ensures high accuracy with a root mean square error of 0.39%. Furthermore, the Phased-RUL Approach attains high precision across the battery’s full lifecycle with the minimum mean absolute error of 3.7 cycles (1.38% of the full lifecycle). It shows superior accuracy in the second phase with an error below 2 cycles and achieves particularly reliable EOL prediction with a maximum error of one cycle.</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

State of health and remaining useful life full lifecycle prediction for lithium-ion battery based on frequency domain interpolation and phased approach

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Keyi He,
  • Xilin Dai,
  • Xu Li,
  • Ruidi Zhou,
  • Fanfan Lin,
  • Weifeng Zhang,
  • Hao Ma

摘要

State-of-health (SOH) estimation and remaining useful life (RUL) prediction of lithium-ion batteries are of great importance in ensuring safe and reliable management of electric vehicles. However, due to the phenomena of capacity regeneration and nonlinear degradation, SOH estimation faces challenges in achieving high precision while maintaining a low model parameter count. Additionally, accurate prediction of RUL throughout the battery’s entire lifecycle is a challenge. To solve the problems, FISNet is proposed to predict SOH and the Phased-RUL Approach is proposed to estimate RUL. FISNet employs a complex-valued linear layer to learn the periodic patterns of the SOH curve and perform reasonable interpolation, specifically capturing degradation trends at low frequencies and regeneration phenomena at high frequencies. The Phased-RUL Approach employs different techniques in two phases of the battery to achieve high precision. In the first phase, it employs a data-driven approach to directly predict RUL through a hybrid model integrating convolutional neural networks, Box-Cox transformation, and multi-layer perceptron block. In the second phase, it employs FISNet to predict SOH and calculate RUL. The experimental results demonstrate that FISNet significantly reduces the model parameter count to 1/250 of the average of four current models and ensures high accuracy with a root mean square error of 0.39%. Furthermore, the Phased-RUL Approach attains high precision across the battery’s full lifecycle with the minimum mean absolute error of 3.7 cycles (1.38% of the full lifecycle). It shows superior accuracy in the second phase with an error below 2 cycles and achieves particularly reliable EOL prediction with a maximum error of one cycle.