摘要
针对光伏组件红外热图像故障检测问题,提出一种改进的DenseNet模型——DenseNet-46模型,改善了原本DenseNet模型网络参数量大、特征通道数目过多、有效特征不明显的问题。利用DenseNet-46模型对光伏组件红外热图像进行特征提取分析,使用J48算法从提取的特征中进行特征选择,并对DenseNet-46、DenseNet-121、ResNet-50、VGG-19这4种模型在光伏组件故障诊断中的准确率进行对比分析。研究结果表明:J48算法在特征选择阶段具有高效性,可实现96.82%的高分类精准度;DenseNet-46模型的光伏组件故障识别准确率可达99.2%,显著优于其他3种模型,具有良好的光伏组件故障诊断性能。DenseNet-46模型不仅降低了模型复杂性,还减少了特征信息损失,为光伏组件故障检测提供了一种高效、准确的解决方案,有望推动光伏发电系统的智能化运维发展。
Aiming at the fault detection problem of PV module infrared thermal images,this paper proposes an improved DenseNet model,DenseNet-46 model,which addresses the issues of large network parameters,excessive feature channels,and effective features less prominent in the original DenseNet model.The DenseNet-46 model is used for feature extraction and analysis of PV module infrared thermal images,and the J48 algorithm is employed for feature selection from the extracted features.The accuracy of four models,namely DenseNet-46,DenseNet-121,ResNet-50,and VGG-19,is compared analysis in faults diagnosing in PV modules.The research results show that the J48 algorithm achieves high efficiency in feature selection,with a classification accuracy of 96.82%.The DenseNet-46 model achieves a recognition accuracy of PV module fault of 99.2%,significantly outperforming the other three models and demonstrating excellent performance in PV module fault diagnosis.DenseNet-46 model not only reduces model complexity but also minimizes feature information loss,providing an efficient and accurate solution for PV module fault detection,and contributing to the intelligent operation and maintenance of PV power generation systems.
作者
何婷
乔俊强
He Ting;Qiao Junqiang(Gansu Natural Energy Research Institute,Lanzhou 730046,China;Gansu Key Laboratory of Solar Energy Utilization,Lanzhou 730046,China)
出处
《太阳能》
2025年第8期45-53,共9页
Solar Energy
基金
甘肃自然能源研究所软科学专项项目(2024SR-02)
甘肃省自然科学基金:项目(23JRRA1664)。
