Scholarlink

<p>Developing dependable diagnostic solutions for battery-powered systems remains a critical challenge owing to the dynamic operating conditions, signal inconsistencies, and skewed class distributions.Conventional deep-learning models often struggle with these issues, resulting in reduced performance and adaptability.To address these obstacles, this research introduces a comprehensive diagnostic framework that integrates multiscale signal analysis, spiking graph-based modeling, and adaptive hyperparameter optimization.The methodology starts with a structured preprocessing phase that addresses missing values using median and forward-fill strategies, detects and eliminates outliers with statistical filters, and applies normalization to ensure consistent feature scaling.Temporal and spectral patterns are captured using the Wavelet Scattering Transform (WST), which offers robust signal representation. These representations are structured into graph sequences and processed through a Spiking Graph Transformer Network (SGTN), which fuses attention-driven spatiotemporal learning with event-based neural computations. To fine-tune model parameters, the Adaptive Arctic Wolf Optimization Algorithm (AAWOA) is employed, enhancing convergence efficiency and generalization. Experimental evaluations demonstrated strong results, with the model achieving 99.2% classification accuracy, 99.5% precision, and an energy consumption of just 0.03836&#xa0;J per prediction.The regression metrics also showed high fidelity, with an MAE of 0.09, MSE of 0.018, and RMSE of 0.1340, whereas the framework remained resilient to test-time variability, exhibiting only a 0.5% performance drop.These findings highlight the potential of the system for real-time, accurate, and energy-efficient battery health assessments across various operational scenarios.</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

An Adaptive Battery Health Monitoring Framework Using Wavelet Scattering and Spiking Graph Transformers Optimized by Arctic Wolf Algorithm

  • 期刊论文

  • 发表日期：

    2025年8月6日

查看全文
  • Ramamoorthy ML.,
  • Selvaperumal S.,
  • Nagarajan R.

摘要

Developing dependable diagnostic solutions for battery-powered systems remains a critical challenge owing to the dynamic operating conditions, signal inconsistencies, and skewed class distributions.Conventional deep-learning models often struggle with these issues, resulting in reduced performance and adaptability.To address these obstacles, this research introduces a comprehensive diagnostic framework that integrates multiscale signal analysis, spiking graph-based modeling, and adaptive hyperparameter optimization.The methodology starts with a structured preprocessing phase that addresses missing values using median and forward-fill strategies, detects and eliminates outliers with statistical filters, and applies normalization to ensure consistent feature scaling.Temporal and spectral patterns are captured using the Wavelet Scattering Transform (WST), which offers robust signal representation. These representations are structured into graph sequences and processed through a Spiking Graph Transformer Network (SGTN), which fuses attention-driven spatiotemporal learning with event-based neural computations. To fine-tune model parameters, the Adaptive Arctic Wolf Optimization Algorithm (AAWOA) is employed, enhancing convergence efficiency and generalization. Experimental evaluations demonstrated strong results, with the model achieving 99.2% classification accuracy, 99.5% precision, and an energy consumption of just 0.03836 J per prediction.The regression metrics also showed high fidelity, with an MAE of 0.09, MSE of 0.018, and RMSE of 0.1340, whereas the framework remained resilient to test-time variability, exhibiting only a 0.5% performance drop.These findings highlight the potential of the system for real-time, accurate, and energy-efficient battery health assessments across various operational scenarios.