A Simple and Effective KAN-Based Architecture for Accurate Battery RUL Prediction

A Simple and Effective KAN-Based Architecture for Accurate Battery RUL Prediction

Accurately estimating a lithium-ion battery’s Remaining Useful Life (RUL) is crucial for ensuring the safety and reliability of battery management systems. However, the performance of emerging architectures, such as Kolmogorov-Arnold Networks (KANs), is often hindered by the significant n...

Full description

Saved in:
Bibliographic Details
Main Authors: Guangzai Ye, Li Feng, Jianlan Guo, Yuqiang Chen, Shufei Li
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Remaining useful life prediction
multivariate time series
Kolmogorov-Arnold networks
channel independent
variational mode decomposition
lithium-ion batteries
Online Access:https://ieeexplore.ieee.org/document/11084810/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Accurately estimating a lithium-ion battery’s Remaining Useful Life (RUL) is crucial for ensuring the safety and reliability of battery management systems. However, the performance of emerging architectures, such as Kolmogorov-Arnold Networks (KANs), is often hindered by the significant noise and complex temporal dynamics present in real-world battery data. To overcome this, we propose a novel Knowledge Distillation-Based Denoising and Channel-Independent Kolmogorov-Arnold Networks (DCI-KANs) architecture, specifically designed to enhance robustness and accuracy for multivariate time series RUL prediction. Our approach integrates a VMD-based denoising mechanism, compressed into a more compact KAN model via knowledge distillation, to mitigate noise efficiently. It also incorporates a channel-independent KANs structure with a regularized weighted loss function to handle variable-specific degradation patterns. Experimental results on two public battery datasets show that DCI-KANs substantially outperform existing state-of-the-art methods in RUL prediction accuracy, achieving a 43 % and 22 % reduction in Absolute Relative Error (ARE) on the NASA and CALCE datasets, respectively. This work presents a simple yet effective framework that makes KANs a practical, robust, and computationally efficient solution for the challenging task of real-world battery RUL prediction.
ISSN:2169-3536

Similar Items