Scholarlink

<p>Reliable fault detection in lithium-ion (Li-ion) batteries is essential for safety and lifetime management in electric vehicles (EVs); however, existing methods often suffer from noise sensitivity, limited fault generalization, and insufficient robustness under varying operating conditions. This paper proposes a hybrid Spike-Induced Graph Neural Network integrated with Orangutan Optimization Algorithm (SIGNN-OOA) to improve classification accuracy and stability under realistic EV operating environments. Maximum Correntropy Unbiased Minimum-Variance Filtering (MCUMVF) was used to pre-process the real EV battery data over 18&#xa0;months and Spatio-Temporal Embedding Fusion Transformer (STEFT) was used to extract key spatio-temporal statistics features. SIGNN is then used to predict internal and external battery faults, while OOA optimizes network parameters to avoid local minima and enhance convergence. Experimental analysis shows that the proposed method has 98.6% accuracy with internal faults, and 98.8% with external faults, with Root Mean Squared Error (RMSE) of 0.16, Mean Absolute Percentage Error (MAPE) of 5.31 which is better than Deep Learning (DL), Genetic Algorithm-optimized Extreme Learning Machine (GA-ELM), Convolutional Neural Network Long Short-Term Memory (CNN-LSTM), Bidirectional Long Short-Term Memory Neural Network (Bi-LSTMNN), and Broad Belief Network (BBN). These findings show that the suggested SIGNN-OOA framework presents an effective and scalable solution to the diagnosis of EV battery fault.</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

Intelligent fault detection and classification of lithium-ion batteries using spike-induced graph neural network and orangutan optimization algorithm framework

  • 期刊论文

  • 发表日期：

    2026年2月19日

查看全文
  • Allam Balaram,
  • V. Ramu,
  • L. Guganathan,
  • J. Anupama

摘要

Reliable fault detection in lithium-ion (Li-ion) batteries is essential for safety and lifetime management in electric vehicles (EVs); however, existing methods often suffer from noise sensitivity, limited fault generalization, and insufficient robustness under varying operating conditions. This paper proposes a hybrid Spike-Induced Graph Neural Network integrated with Orangutan Optimization Algorithm (SIGNN-OOA) to improve classification accuracy and stability under realistic EV operating environments. Maximum Correntropy Unbiased Minimum-Variance Filtering (MCUMVF) was used to pre-process the real EV battery data over 18 months and Spatio-Temporal Embedding Fusion Transformer (STEFT) was used to extract key spatio-temporal statistics features. SIGNN is then used to predict internal and external battery faults, while OOA optimizes network parameters to avoid local minima and enhance convergence. Experimental analysis shows that the proposed method has 98.6% accuracy with internal faults, and 98.8% with external faults, with Root Mean Squared Error (RMSE) of 0.16, Mean Absolute Percentage Error (MAPE) of 5.31 which is better than Deep Learning (DL), Genetic Algorithm-optimized Extreme Learning Machine (GA-ELM), Convolutional Neural Network Long Short-Term Memory (CNN-LSTM), Bidirectional Long Short-Term Memory Neural Network (Bi-LSTMNN), and Broad Belief Network (BBN). These findings show that the suggested SIGNN-OOA framework presents an effective and scalable solution to the diagnosis of EV battery fault.