Scholarlink

<p>This manuscript introduces two Adaptive Neuro-Fuzzy Inference Systems developed to predict the energy output of Photovoltaic cells. These models are trained using Electroluminescence imagery of the cells for input data along their Current–Voltage curves, which offer insights output power of cells. The input characteristics of the cells are quantified based on pixel distribution and classified into three distinct categories: Black, White, and Gray values. The second model enhances this representation by incorporating an additional fuzzy categorization input, derived from a Mamdani Classifier Fuzzy Logic Model. By combining the rule-based interpretability of Fuzzy Logic with the adaptive learning capabilities of Artificial Neural Networks, the Adaptive Neuro-Fuzzy Inference System (ANFIS) emerges as an alternative to Convolutional Neural Networks (CNNs). This approach contributes to Explainable Artificial Intelligence by addressing one of the major limitations of CNNs—the lack of symbolic knowledge representation, while maintaining robust learning performance. Comparative analysis with other Machine Learning techniques demonstrates the enhanced performance provided by ANFIS models, achieving a Mean Absolute Error (MAE) of 0.053 and a Mean Squared Error (MSE) of 0.007.</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

ANFIS-based output power estimation in photovoltaic cells using electroluminescence image features

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Hector Felipe Mateo-Romero,
  • Mario Eduardo Carbonó de la Rosa,
  • Luis Hernández-Callejo,
  • Miguel Ángel González-Rebollo,
  • Valentín Cardeñoso-Payo,
  • Victor Alonso-Gómez,
  • Oscar Martínez-Sacristán,
  • Sara Gallardo-Saavedra,
  • Adalberto José Opsino Castro

摘要

This manuscript introduces two Adaptive Neuro-Fuzzy Inference Systems developed to predict the energy output of Photovoltaic cells. These models are trained using Electroluminescence imagery of the cells for input data along their Current–Voltage curves, which offer insights output power of cells. The input characteristics of the cells are quantified based on pixel distribution and classified into three distinct categories: Black, White, and Gray values. The second model enhances this representation by incorporating an additional fuzzy categorization input, derived from a Mamdani Classifier Fuzzy Logic Model. By combining the rule-based interpretability of Fuzzy Logic with the adaptive learning capabilities of Artificial Neural Networks, the Adaptive Neuro-Fuzzy Inference System (ANFIS) emerges as an alternative to Convolutional Neural Networks (CNNs). This approach contributes to Explainable Artificial Intelligence by addressing one of the major limitations of CNNs—the lack of symbolic knowledge representation, while maintaining robust learning performance. Comparative analysis with other Machine Learning techniques demonstrates the enhanced performance provided by ANFIS models, achieving a Mean Absolute Error (MAE) of 0.053 and a Mean Squared Error (MSE) of 0.007.