针对光伏组件中红外热斑的识别及分类需训练样本数量较大以及准确率还有待提高的问题,提出一种基于持续同调算法与卷积神经网络相结合的热斑识别方法。首先使用拓扑数据分析中的持续同调算法,将红外热图像中RGB三通道上的数值映射到三维坐标系形成三维点云,然后进行持续同调计算,预先提取出图片内部所包含的拓扑特征,再将提取出的特征向量化处理后以固定的顺序排列,映射到图像的像素中去,并与图片的亮度及对比度特征相结合,最后将处理后的图像数据输入到调整后的LeNet-5卷积神经网络模型中,实现对光伏红外热斑的分类识别,并通过混淆矩阵计算各项性能指标,以评估模型的性能。实验结果表明,该模型有效地提取出隐藏在图像内部的高维拓扑特征,并与其他特征进行有利地互补结合,解决图像数据无法直接输入到持续同调算法中以及高维度拓扑特征无法直接作为深度学习模型输入的问题,同时提高了光伏红外热斑的分类识别准确率,且显著减少了所需的计算资源。
Abstract
Aiming at the problem that the identification and classification of infrared hot spots in photovoltaic modules require a large number of training samples and the accuracy needs to be improved, a hot spot recognition method based on the combination of persistent homology algorithm and convolutional neural network is proposed. The persistent homology algorithm in topological data analysis is first used to map the values on the RGB three channels in the infrared thermal image to the three-dimensional coordinate system to form a three-dimensional point cloud, and then the persistent homology calculation is carried out to extract the topological features contained in the picture in advance, and then the extracted features are vectorized and arranged in a fixed order, mapped to the pixels of the image, and combined with the brightness and contrast characteristics of the picture. Finally, the processed image data is input into the adjusted LeNet-5 convolutional neural network model to realize the classification and identification of photovoltaic infrared hot spots, and various performance indicators are calculated through the confusion matrix to evaluate the performance of the model. Experimental results show that the model effectively extracts the high-dimensional topological features hidden in the image, and complements and combines them with other features favorably, which solves the problem that the image data cannot be directly input into the persistent homology algorithm and the high-dimensional topological features cannot be directly used as input to the deep learning model. Additionally, the classification and recognition accuracy of photovoltaic infrared hot spots is improved, and the computing resources required are significantly reduced.
关键词
光伏组件 /
特征提取 /
卷积神经网络 /
拓扑数据分析 /
持续同调 /
光伏热斑
Key words
PV modules /
feature extraction /
convolutional neural network /
topological data analysis /
persistent homology /
hot spot
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] QIAN X Y, BAI Y, HUANG W L, et al.Fuzzy technique application in selecting photovoltaic energy and solar thermal energy production in Belt and Road countries[J]. Journal of energy storage, 2021, 41: 102865.
[2] 孙海蓉, 李帆. 基于注意力机制的光伏热斑识别[J]. 太阳能学报, 2023, 44(2): 453-459.
SUN H R, LI F.Photovoltaic hot spot recognition based on attention mechanism[J]. Acta energiae solaris sinica, 2023, 44(2): 453-459.
[3] 孙海蓉, 李号. 基于深度迁移学习的小样本光伏热斑识别方法[J]. 太阳能学报, 2022, 43(1): 406-411.
SUN H R, LI H.Photovoltaic hot spot identification method for small sample based on deep transfer learning[J]. Acta energiae solaris sinica, 2022, 43(1): 406-411.
[4] 车曦. 基于红外图像识别的光伏组件热斑故障检测方法研究[D]. 重庆: 重庆大学, 2015.
CHE X.Research on hot spot fault detection method of photovoltaic module based on infrared image recognition[D]. Chongqing: Chongqing University, 2015.
[5] 蔺怡, 欧阳名三, 汪义鹏, 等. 基于主成分分析的光伏热斑红外图像混合噪声去噪方法[J]. 重庆工商大学学报(自然科学版), 2024, 41(1): 30-37.
LIN Y, OUYANG M S, WANG Y P, et al.Denoising method of mixed noise in photovoltaic hot spot infrared images based on principal component analysis[J]. Journal of Chongqing Technology and Business University (natural science edition), 2024, 41(1): 30-37.
[6] BUKKURI A, ANDOR N, DARCY I K.Applications of topological data analysis in oncology[J]. Frontiers in artificial intelligence, 2021, 4: 659037.
[7] LIN B H.Topological data analysis in time series: temporal filtration and application to single-cell genomics[J]. Algorithms, 2022, 15(10): 371.
[8] 赵彦. 拓扑学的同调方法及其应用[D]. 石家庄: 河北师范大学, 2021.
ZHAO Y.Homology method of topology and its application[D]. Shijiazhuang: Hebei Normal University, 2021.
[9] EDWARDS P, SKRUBER K, MILIĆEVIĆ N, et al. TDAExplore: quantitative analysis of fluorescence microscopy images through topology-based machine learning[J]. Patterns, 2021, 2(11): 100367.
[10] 孙海蓉, 周映杰. 基于散点图-AlexNet网络的光伏红外热图像识别方法[J]. 太阳能学报, 2023, 44(1): 55-61.
SUN H R, ZHOU Y J.Photovoltaic infrared thermal image recognition method based on scatter plot-AlexNet network[J]. Acta energiae solaris sinica, 2023, 44(1): 55-61.
[11] 王洪申, 李昌德. 基于LeNet-5的试卷手写分数无分割识别方法[J]. 计算机仿真, 2023, 40(2): 459-465.
WANG H S, LI C D.An undivided recognition method for handwritten score of test paper based on LeNet-5[J]. Computer simulation, 2023, 40(2): 459-465.
[12] KOULALI I, ESKIL M T.Unsupervised textile defect detection using convolutional neural networks[J]. Applied soft computing, 2021, 113: 107913.
[13] MANSOUR R F, AL-MARGHILNAI A.Glaucoma detection using novel perceptron based convolutional multi-layer neural network classification[J]. Multidimensional systems and signal processing, 2021, 32(4): 1217-1235.
[14] 孙海蓉, 潘子杰, 晏勇. 基于深度卷积自编码网络的小样本光伏热斑识别与定位[J]. 华北电力大学学报(自然科学版), 2021, 48(4): 91-98.
SUN H R, PAN Z J, YAN Y.Identification and location of photovoltaic hot spots with small samples based on deep convolutional autoencoder network[J]. Journal of North China Electric Power University(natural science edition), 2021, 48(4): 91-98.
[15] 孙海蓉, 周映杰, 张镇韬, 等. 基于改进自私羊群算法的光伏红外热图像热斑识别方法[J]. 中国电机工程学报, 2022, 42(24): 8942-8951.
SUN H R, ZHOU Y J, ZHANG Z T, et al.Hot spot recognition method of photovoltaic infrared thermal image based on improved selfish herd algorithm[J]. Proceedings of the CSEE, 2022, 42(24): 8942-8951.
PDF(2592 KB)
