关键词: 容量增量, 变分模态双重分解, 噪声, 主退化趋势, 波动趋势, 长短期记忆网络

Abstract:

To address the issue of inaccurate battery health state estimation caused by strong interference in the lithium battery capacity increment curve, such as sensor measurement noise or varying operational conditions, this paper proposes an innovative solution using triple variational mode decomposition (VMD) to improve the accuracy of state of health (SOH) estimation. First, a dual VMD technique is utilized to denoise the distorted battery capacity increment curves. These interference sources include global voltage noise, local voltage noise, and local current mutations. Peak features were then extracted from the denoised curves. To further enhance the ability of these peak features to represent the battery's health state, a second VMD decomposition is applied to the extracted peak features. Using Pearson correlation analysis, the mode components are reconstructed into two sub-components: the main degradation trend that reflects the overall attenuation of the feature over time, and the fluctuation trend that captures short-term variations in the feature. These two components are used together as health indicators for SOH estimation. Finally, based on the NASA dataset, battery SOH estimation validation experiments were conducted using algorithms such as long short-term memory networks. The experimental results show that the proposed method effectively estimates the lithium-ion battery SOH under strong interference conditions, achieving high estimation accuracy and demonstrating significant advantages.

Key words: capacity increment, variational mode decomposition, noise, primary degradation trend, fluctuation trend, long short term memory network