Scholarlink

Background <p>The prediction of Remaining Useful Life (RUL) and capacity fade of lithium-ion batteries (LIBs) is a very challenging task because battery degradation processes are highly nonlinear, and because of the weaknesses of traditional predictive models. The applications of the LIBs include EVs, energy storage, and portable electronics, and LIBs are not only necessary in these applications but also degrade with time, so there is a need to ensure reliable health monitoring and predictive maintenance.</p> Objectives <p>The major aim of the proposed research is to develop a powerful and dependable model for accurately forecasting the RUL and capacity fade of lithium-ion batteries. The aim of this framework is proactive maintenance and extends battery lifespan by capturing complex degradation patterns.</p> Methodology <p>The proposed approach integrates the Variational Relation Vector Graph Neural Networks (VRVGNN) and the Starfish Optimization Algorithm (SOA). Battery data are preprocessed using the Information Exchange Multi-Bernoulli Filter (IEMBF) and key features are selected using Leaf in Wind Optimization (LWO). The VRVGNN models complicated mutual dependencies between battery characteristics, and SOA continuously updates the weight factors of the network to achieve better predictive accuracy. The framework is validated using a dataset containing capacity, voltage, cycle count, and temperature parameters.</p> Findings <p>Results demonstrate that the VRVGNN-SOA model has a 98% of prediction accuracy, RMSE of 0.014, and R<sup>2</sup> of 99 and is better to the existing techniques such as DCNN, RF-XGBoost and CNN-BO. The proposed model excellently captures nonlinear degradation trends, providing more dependable RUL and capacity fade predictions, thus supporting enhanced battery management and extending operational lifespan.</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

Sustainable Lithium-Ion Battery Operation Using Variational Relation Vector Graph Neural Networks for High-Fidelity Prediction of Capacity Fade and Remaining Useful Life

  • 期刊论文

  • 发表日期：

    2026年3月11日

查看全文
  • K. Manimala,
  • A. M. Kalpana

摘要

Background

The prediction of Remaining Useful Life (RUL) and capacity fade of lithium-ion batteries (LIBs) is a very challenging task because battery degradation processes are highly nonlinear, and because of the weaknesses of traditional predictive models. The applications of the LIBs include EVs, energy storage, and portable electronics, and LIBs are not only necessary in these applications but also degrade with time, so there is a need to ensure reliable health monitoring and predictive maintenance.

Objectives

The major aim of the proposed research is to develop a powerful and dependable model for accurately forecasting the RUL and capacity fade of lithium-ion batteries. The aim of this framework is proactive maintenance and extends battery lifespan by capturing complex degradation patterns.

Methodology

The proposed approach integrates the Variational Relation Vector Graph Neural Networks (VRVGNN) and the Starfish Optimization Algorithm (SOA). Battery data are preprocessed using the Information Exchange Multi-Bernoulli Filter (IEMBF) and key features are selected using Leaf in Wind Optimization (LWO). The VRVGNN models complicated mutual dependencies between battery characteristics, and SOA continuously updates the weight factors of the network to achieve better predictive accuracy. The framework is validated using a dataset containing capacity, voltage, cycle count, and temperature parameters.

Findings

Results demonstrate that the VRVGNN-SOA model has a 98% of prediction accuracy, RMSE of 0.014, and R2 of 99 and is better to the existing techniques such as DCNN, RF-XGBoost and CNN-BO. The proposed model excellently captures nonlinear degradation trends, providing more dependable RUL and capacity fade predictions, thus supporting enhanced battery management and extending operational lifespan.