Scholarlink

<p>Power quality degradation in distribution systems necessitates a solid fault diagnosis of Unified Power Quality Conditioners (UPQCs) that are susceptible to power-switch, sensor, and control-system failures. Conventional model-based and machine learning methods have low adaptability, interpretability, and prediction ability. We introduce AIMARSE, a new Deep Learning-enhanced multi-agent architecture that combines hierarchical feature extraction based on Convolutional Neural Networks (CNNs) with multi-agent reinforcement learning to make adaptive decisions. The architecture is based on five expert diagnostic agents, deployed in a cooperative learning environment with self-evolution, enabling the system to continually improve performance without retraining. It has been experimentally verified on a 25 kVA three-phase UPQC simulation platform, with much better performance: 97.8% fault detection accuracy, 94.6% multi-class fault classification accuracy with six fault configurations, and 5.2&#xa0;ms average detection latency. The structure demonstrates 120&#xa0;s early-fault prognostics, which is much better than the reactive strategies. The statistical significance test confirms strong improvements (<i>p</i> &lt; 0.001, Cohen’s d &gt; 0.8) across all metrics. The integration of explainable AI using SHAP and LIME can provide transparency in diagnostic results, with 94% acceptance among operators. The suggested AIMARSE paradigm advances smart predictive maintenance for power-quality equipment and helps create a self-healing smart grid system.</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

Deep learning-enhanced AIMARSE for predictive fault diagnosis in power quality systems

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Dinesh Kumar Nishad,
  • A. N. Tiwari,
  • Saifullah Khalid

摘要

Power quality degradation in distribution systems necessitates a solid fault diagnosis of Unified Power Quality Conditioners (UPQCs) that are susceptible to power-switch, sensor, and control-system failures. Conventional model-based and machine learning methods have low adaptability, interpretability, and prediction ability. We introduce AIMARSE, a new Deep Learning-enhanced multi-agent architecture that combines hierarchical feature extraction based on Convolutional Neural Networks (CNNs) with multi-agent reinforcement learning to make adaptive decisions. The architecture is based on five expert diagnostic agents, deployed in a cooperative learning environment with self-evolution, enabling the system to continually improve performance without retraining. It has been experimentally verified on a 25 kVA three-phase UPQC simulation platform, with much better performance: 97.8% fault detection accuracy, 94.6% multi-class fault classification accuracy with six fault configurations, and 5.2 ms average detection latency. The structure demonstrates 120 s early-fault prognostics, which is much better than the reactive strategies. The statistical significance test confirms strong improvements (p < 0.001, Cohen’s d > 0.8) across all metrics. The integration of explainable AI using SHAP and LIME can provide transparency in diagnostic results, with 94% acceptance among operators. The suggested AIMARSE paradigm advances smart predictive maintenance for power-quality equipment and helps create a self-healing smart grid system.